CN109671103A - Method for tracking target and device - Google Patents
Method for tracking target and device Download PDFInfo
- Publication number
- CN109671103A CN109671103A CN201811516817.9A CN201811516817A CN109671103A CN 109671103 A CN109671103 A CN 109671103A CN 201811516817 A CN201811516817 A CN 201811516817A CN 109671103 A CN109671103 A CN 109671103A
- Authority
- CN
- China
- Prior art keywords
- target
- frame
- tracking
- score
- goal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
- G06T7/251—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
During the present invention is in order to solve in object tracking process especially multiple target tracking, algorithm is complicated, computationally intensive problem, provide a kind of method for tracking target based on tracker, feature extractor and goal-based assessment device, include: acquisition picture frame, obtains position score using the target tracker；According to the position of the target in previous frame image and size, sliding forms frame to be detected in object detection area, calculate the similarity between the feature vector and template characteristic vector of the frame to be detected, taking out the highest threshold value of score frame to be detected is detection block, and the destination probability score of the detection block is obtained using goal-based assessment device；When judging that soprano is greater than threshold value in the position score and the destination probability score, identification is tracked successfully, and using corresponding frame as target.Meanwhile providing corresponding device.Method for tracking target and device in the embodiment of the present invention can be quick and uncertain realization target following.
Description
Technical field
The invention belongs to technical field of image processing, in particular to a kind of method for tracking target, tracking device, meter
Calculation machine readable medium and electronic equipment.
Background technique
Currently, in target identification, object detection field achieves breakthrough due to the fast development of neural network,
But it is but developed slowly in target tracking domain.The reason is that neural network need it is powerful out with a large amount of diversity sample competence exertions
Capability of fitting, but target following is only marked given first frame, and training sample is single；Simultaneously target following to algorithm when
Effect property has higher requirement, and the general calculation amount of neural network is larger, it is difficult in real time.The present invention realizes a kind of based on neural network
Real-time multi-target long-time tracking.The technology can be used for smart television, PC, tablet computer and smart phone
Equal electronic equipments.
Existing target long-time tracking is mainly TLD (Tracking-Learning-Detection) series, still
This method effect is general, more sensitive to environmental change, target surface variation, is easily lost；It is computationally intensive simultaneously, it cannot achieve
Multiple target tracking requirement.
Tracking facing challenges are also huge to multiple target for a long time in actual scene:
(1) environment is complicated, and light, motion blur, angle change are blocked
(2) sample is single, it is difficult to good detector and classifier at training
(3) computationally intensive
(4) after target is lost, need recapture to target
Thus how to develop quick and stable target tracking algorism has become current urgent problem, and for more
It for target tracking algorism, tracks, detect again, algorithm optimization is vital part.
Summary of the invention
The present invention in order to solve the disadvantage that the above-mentioned prior art, propose a kind of quick and stable method for tracking target and
Device can efficiently apply in single goal and multiple target tracking scene, specifically, in a first aspect, the embodiment of the present invention mentions
A kind of method for tracking target has been supplied, has been included the following steps:
S120, picture frame is obtained, the current location rectangle of target is obtained using target tracker, and to current location rectangle
Feature vector is extracted, the position score of current location rectangle is judged using goal-based assessment device；
S130, the position according to the target in previous frame image and size, are extended to form mesh in current frame image
Mark detection zone, sliding forms frame to be detected in the object detection area, calculate the feature vector of the frame to be detected with
Similarity between template characteristic vector, taking out the highest threshold value of score frame to be detected is detection block, utilizes goal-based assessment device
Obtain the destination probability score of the detection block；
S140, when judging in the position score and the destination probability score that soprano is greater than threshold value, assert tracking at
Function, and using the image in the corresponding detection block of top score or current location rectangle as target.
Further, the step S140 further includes judging highest in the position score and the destination probability score
When person is less than threshold value, tracking failure is assert；In the step S130, the position according to the target in previous frame image and big
Small, being extended in current frame image and forming object detection area includes: when judging the failure of previous frame image trace, to set institute
State the whole region that object detection area is current frame image.
Further, it further comprises the steps of:
When tracking successfully in S150, judgment step S140, according to the target feature vector and current mould in current frame image
Plate features vector updates the template characteristic vector, repeats step S120 to step S140 until image terminates.
Further, the step of update template characteristic vector includes: to be updated according to the following formula,
FeatTmp=featTmp × (1-alpha)+featCur × alpha, wherein featTmp is template characteristic vector, featCur
For present frame target feature vector, alpha is learning rate.
Further, it further comprises the steps of:
When tracking successfully in S160, judgment step S140, the feature vector for calculating current goal is commented with for training objective
The target is added to by the similarity for estimating the feature vector of the sample in the sample pool of device when judging that similarity is less than threshold value
In the sample pool, updated goal-based assessment device is formed using the training of updated sample pool, is commented using updated target
Device is estimated as the goal-based assessment device, repeats step S120 to step S140 until image terminates.
Further, further includes:
S110, tracking target frame is obtained from initial frame image, according to the tracking target frame initialized target tracker
With goal-based assessment device, the feature vector in the tracking target frame is extracted as template characteristic vector；
It is wrapped in the step S110 according to the step of tracking target frame initialized target tracker and goal-based assessment device
It includes: regression training being carried out according to the target following frame and initial frame image, obtains the target tracker.
Further, according to the tracking target frame initialized target tracker and goal-based assessment device in the step S110
The step of include: to generate to calculate with machine frame described with machine frame and the friendship of the tracking target frame and ratio in initial frame image, sentence
Break the friendship and when than being greater than first threshold, will it is corresponding with machine frame as positive sample, judge described hand over and ratio is less than the second threshold
When value, by corresponding with machine frame negative sample the most, the goal-based assessment is formed using the positive sample and negative sample training
Device.
Second aspect of the embodiment of the present invention provides a kind of target tracker, comprising:
Position score obtains module, and for obtaining picture frame, the current location of target is obtained using the target tracker
Rectangle, and to current location rectangular extraction feature vector, the position score of current location rectangle is judged using goal-based assessment device；
Destination probability obtain module, according to the position of the target in previous frame image and size, in current frame image into
Row extension forms object detection area, and sliding forms frame to be detected in the object detection area, calculates the frame to be detected
Feature vector and template characteristic vector between similarity, taking out the highest threshold value of score frame to be detected is detection block, benefit
The destination probability score of the detection block is obtained with goal-based assessment device；
Target Acquisition module, for judging that soprano is greater than threshold value in the position score and the destination probability score
When, identification tracks successfully, and using the image in the corresponding detection block of top score or current location rectangle as target.
Third aspect present invention provides a kind of electronic equipment, comprising:
One or more processors；
Storage device is stored thereon with one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processing
Device realizes any of the above-described method.
Fourth aspect present invention provides a kind of computer-readable medium, is stored thereon with computer program, wherein institute
It states and realizes any of the above-described method when program is executed by processor.
The embodiment of the present invention is by the way that using initial frame image and the initial target frame that marks in advance, then progress tracker is commented
Estimate the initialization training of device and form original template feature, later, target position frame in subsequent frame image is obtained according to tracker,
The first probability is obtained using evaluator, then according to sliding retrieval is extended in the region of previous frame target detection, judges phase
Like property, the corresponding characteristic use evaluator of preceding several higher frames of similitude is taken out, corresponding second probability is calculated, judges that first is general
When probability corresponding to highest frame is greater than threshold value in rate and the second probability, judgement is tracked successfully.In the embodiment of the present invention with
Track method calculates quickly, and under multiple target scene, effect is obvious.Under multiple target scene, initial target frame has multiple.It is corresponding
Ground, subsequent result again more to be covered, so as to determine multiple tracking targets.
Detailed description of the invention
The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, and attached drawing is schematically without that should manage
Solution is carries out any restrictions to the present invention, in the accompanying drawings:
Fig. 1 is the system architecture schematic diagram that method for tracking target, the extraction element in some examples of the present invention are run；
Fig. 2 is the method for tracking target flow chart in some examples of the present invention；
Fig. 3 is method for tracking target feature extractor and the signal of goal-based assessment device network in some embodiments of the invention
Figure；
Fig. 4 is the method for tracking target flow diagram in other embodiments of the invention；
Fig. 5 is the method for tracking target flow diagram in other embodiments of the invention；
Fig. 6 is the method flow schematic diagram in other embodiments of the invention in method for tracking target；
Fig. 7 is the target tracker system architecture schematic diagram in other embodiments of the invention；
Fig. 8 is that method for tracking target or the Computer Systems Organization of extraction element operation are shown in some embodiments of the invention
It is intended to.
Specific embodiment
To better understand the objects, features and advantages of the present invention, with reference to the accompanying drawing and specific real
Applying mode, the present invention is further described in detail.It should be noted that in the absence of conflict, the implementation of the application
Feature in example and embodiment can be combined with each other.
In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention, still, the present invention may be used also
To be implemented using other than the one described here other modes, therefore, protection scope of the present invention is not by described below
Specific embodiment limitation.
Fig. 1 is shown can be using the embodiment of the method for tracking target or view target tracker of the embodiment of the present application
Exemplary system architecture 100.
As shown in Figure 1, system architecture 100 may include terminal device 101,102,103, network 104 and server 105.
Network 104 between terminal device 101,102,103 and server 105 to provide the medium of communication link.Network 104 can be with
Including various connection types, such as wired, wireless communication link or fiber optic cables etc..
User can be used terminal device 101,102,103 and be interacted by network 104 with server 105, to receive or send out
Send data (such as video) etc..Various telecommunication customer end applications, such as video can be installed on terminal device 101,102,103
Playout software, video processing class application, web browser applications, the application of shopping class, searching class application, instant messaging tools, postal
Case client, social platform software etc..
Terminal device 101,102,103 can be hardware, be also possible to software.When terminal device 101,102,103 is hard
When part, the various electronic equipments of data transmission, including but not limited to smart phone, plate are can be with display screen and supported
Computer, pocket computer on knee and desktop computer etc..When terminal device 101,102,103 is software, can install
In above-mentioned cited electronic equipment.Multiple softwares or software module may be implemented into (such as providing distributed clothes in it
The software or software module of business), single software or software module also may be implemented into.It is not specifically limited herein.
Server 105 can be to provide the server of various services, such as to showing on terminal device 101,102,103
Video provides the background server supported.Background server the data such as the image processing requests received such as can analyze
Processing, and processing result (such as the video clip or other data obtained after being split to video) is fed back to and is communicated with
The electronic equipment (such as terminal device) of connection.
It should be noted that method for tracking target provided by the embodiment of the present application can be executed by server 105, accordingly
Ground, target tracker can be set in server 105.In addition, method for tracking target provided by the embodiment of the present application
Can be executed by terminal device 101,102,103, correspondingly, target tracker also can be set in terminal device 101,102,
In 103.
It should be noted that server can be hardware, it is also possible to software.When server is hardware, may be implemented
At the distributed server cluster that multiple servers form, individual server also may be implemented into.It, can when server is software
To be implemented as multiple softwares or software module (such as providing the software of Distributed Services or software module), also may be implemented
At single software or software module.It is not specifically limited herein.
It should be understood that the number of terminal device, network and server in Fig. 1 is only schematical.According to realization need
It wants, can have any number of terminal device, network and server.When the electronic equipment of method for tracking target operation thereon
When not needing to carry out data transmission with other electronic equipments, which can only include running on for dividing video method
Electronic equipment (such as terminal device 101,102,103 or server 105) thereon.
Fig. 2 shows a kind of general flows of method for tracking target according to the embodiment of the present invention, that is, provides a kind of base
In the multiple target long-time tracking of convolutional neural networks.The tracking of multiple targets is identical as a target following mode, more
A target need to only create multiple tracking examples.Fig. 2 shows the general flow figures of the method for the present invention.Main modular and step are such as
Under:
Module introduction:
Basic tracker: the present invention use KCF (KCF Kernelized correlation filter) based on
Track device, this method is for fast with tracking velocity, to strong environmental adaptability, the features such as tracking accuracy is high.Basic tracker: no
It is limited to the high-precisions, quick tracking mode such as KCF tracker, such as MedianFlow tracker tracking.
Feature extractor: with reference to the part feature net in Fig. 3.Input is that image to be processed and feature to be extracted are examined
Frame is surveyed, is exported as the feature vector after the normalization of all detection blocks.Use preceding 3 convolutional layers of VGG16 as feature extraction layer, knot
Closing roi pooling layers and the forward direction operation of normalize layer is after can extract all detection block fixed dimensions normalization
Feature vector.Due to network very little, and by the way of full convolution, therefore it is unrelated with target number to execute number, it is only necessary to
A forward direction operation is executed, speed is quickly.2. feature extractor: be not limited to VGG16 preceding 3 layers of joint roi-pooling and
The mode of normalize, such as ResNet combine roi-pooling and normalize, the modes such as EdgeBoxes.
Goal-based assessment device: with reference to the part validate net in Fig. 3.A sorter network is built with 3 fc layer networks,
Input is characterized the feature vector of the detection block of extractor extraction, exports the probability that the detection block belongs to target.Different target
Network structure is identical, but the weight of network model is different, establishes a goal-based assessment device for each target.Due to each evaluator
Input be feature that feature extractor extracts, and using having 3 layers of fc network of powerful capability of fitting to classify, therefore mesh
The performance for marking evaluator is relatively good；The target frame for needing to assess simultaneously is less, therefore speed is fast.3. goal-based assessment device: being not limited to
Neural network fashion, such as SVM mode.
Key step:
Step1: the target frame groundTruth_box to be tracked marked from initial frame.
Initial situation is that handmarking goes out target to be tracked, and the frame of mark is denoted as: groundTruth_box
(1) basic tracker (KCF tracker) current frame image and groundTruth_box are modeled.The mistake of modeling
Journey is sampled near groundTruth_box, and one recurrence device of training, the recurrence device can calculate the sound of image pixel positions
It should be worth.It is bigger closer to target's center's response, it is smaller further away from target response.
Returning tracker is exactly basis kcf tracker, and training and returning the effect of tracker is the optimal position for finding target
It sets.Then it is assessed with goal-based assessment device, judges whether the target currently tracked is correct.
(2) feature extractor is used to extract the feature vector in groundTruth_box as template characteristic vector
featTmp。
(3) it is generated in initial frame with machine frame, calculates the framed iou value with groundTruth_box of institute, the work of iou > 0.7
For positive sample, iou < 0.3 is used as negative sample, and the diversity of the abundant sample of data enhancing is done to positive sample, and training objective is commented
Estimate device.Wherein iou is framed and groundTruth_box degree of overlapping, and calculation formula is as follows:
Wherein: DetectionRect is detection block, and GroundTruth is target callout box
Iou is to hand over and compare, and is a kind of mode that detection accuracy is measured in target detection.
Data enhancing: being exactly the Random-Rotation that certain angle is done to each positive sample, brightness, contrast, color saturation
Random variation is done, the diversity of sample is increased.
Step1 is algorithm input, and (1), (2), (3) are the target frame initialization algorithms according to handmarking, are to step1
Refinement, (1) is initialization kcf tracker, and (2) are to extract target signature, and (3) are to collect positive negative sample, for training
Goal-based assessment device.Wherein positive sample is the frame handed over and compared with GroundTruth greater than 0.7, and negative sample be friendship and ratio less than 0.3
Frame.
GroundTruth is artificial label target callout box,
Step2: subsequent frame image is read in, the current location tracker_box of target is estimated with KCF；To tracker_
Box extracts feature vector with feature extractor, and the score tracker_score of tracker_box is calculated with goal-based assessment device.
Step3: centered on previous frame target, the range of 1 times of target long side size is respectively extended up and down as mesh
Mark detection zone.Keep original object length-width ratio constant, respectively with the frame of 5 kinds of scale (0.8,0.9,1.0,1.1,1.2) sizes
It is slided in inspection area, sliding stepping is 8, obtains all frames to be detected.To operation before being executed to detection zone, extract each
The feature vector featCurs of detection block, calculate each detection block and template characteristic vector f eatTmp similarity (cos away from
From), highest 5 detection blocks of score are taken out, and assess the probability that this 5 detection zones belong to target with goal-based assessment device
proposal_scores。
(0.8,0.9,1.0,1.1,1.2) it is the multiple proportion of size, is the coefficient that the size of previous frame target is scaled.
It slides stepping: calculating the size of current detection frame according to scale coefficient；Then image horizontal axis, y direction gradually
Mobile detection block, each mobile 8 pixels；Every movement will once obtain a detection zone, be denoted as frame to be detected here.
Forward direction operation: as soon as often obtaining a frame to be detected, it will test the picture material in frame as the defeated of feature extractor
Enter, calculate the output of feature extractor, will be output as the feature of the detection block.Feature extractor is a neural network, forward direction
Operation is exactly given input, is exported according to the weight computing network of network.
The formula for calculating the cos distance between vector is as follows:
Feature vector after being normalization due to what is obtained by feature extractor, the denominator normalization of formula (2)
It is 1, formula (2) is reduced to formula (3), as follows:
Wherein A, B are two groups of vectors, and n is the dimension of vector
Step4: tracker_score and proposal_scores is sorted, takes score soprano as final tracking
As a result tracker_result, if score is less than threshold value (for example, 0.65), then it is assumed that tracking failure.
Step5: occurring again after target is lost, and setting inspection area is full figure, passes through Step3 recapture to mesh
Mark, then the peak response reappeared as target is maximized to emerging goal-based assessment with goal-based assessment device, if most
Big response is greater than threshold value (for example, 0.85), then it is assumed that the success of target recapture.
Step3 recapture mesh calibration method herein is: purpose is to capture target in the current frame, and capture region claims
For area to be tested, there are two types of situations: tracker loses target, and tracker does not lose target.
If tracker does not lose target, the case where being exactly step3, centered on previous frame target, surrounding extension 1
Times target sizes are as area to be tested.Because target will not be run too far between adjacent two frame.
The case where if tracker loses target, is exactly step5, without reference to meaning centered on previous frame target
Justice, so directly setting full figure for area to be tested.
Step6: template renewal is updated when tracking successfully, is not otherwise updated.Update mode are as follows:
FeatTmp=featTmp × (1-alpha)+featCur × alpha (4)
Wherein: featTmp is template characteristic vector, and featCur is present frame target feature vector, and alpha is learning rate
(for example, 0.25).
Step7: goal-based assessment device model modification is updated when tracking successfully, is not otherwise updated.It is collected with k-mean mode
For the sample of different postures as positive sample, the variable quantity of posture is that feature extractor extracts present frame feature vector and all positive samples
The maximum value of the similarity of this feature vector, similarity is bigger, and attitudes vibration is smaller, conversely, attitudes vibration is bigger.If worked as
Previous frame attitudes vibration is larger, then saving the sample is positive sample, and updates current goal evaluator；Otherwise the sample is abandoned,
It does not also update simultaneously.The collection mode of negative sample is full figure uniform sampling, and positive and negative sample proportion control is in 1:3.
K-mean mode collects sample: collecting all samples in time domain into sample pool, the mode of collection uses k-
Mean cluster.It is exactly the complete frame target of every secondary tracking, the similarity of this frame target and the target in sample pool is calculated, if non-
It is often similar, then this frame is abandoned, otherwise, this frame is saved in sample pool.All samples in sample pool are ensured that in this way
This similarity is all very low, and the low diversity for illustrating sample of similarity is all right.(calculating of similarity be calculate present frame sample with
The cos distance of each sampling feature vectors in sample pool)
Attitudes vibration is larger: exactly calculating all samples in present frame and sample pool and calculates similarity, similarity is big, posture
Variation is just small, and otherwise, attitudes vibration is with regard to big.
Re -training goal-based assessment device: goal-based assessment device training sample is trained with the sample in sample pool,
Sample in step1 (3) can participate in training in sample pool.
A kind of multiple target long-time tracking neural network based provided in an embodiment of the present invention, including target with
Track, feature extraction, the parts such as goal-based assessment have the advantage that as follows: using the versatility network model without mark
VGG16 is as feature extractor.Feature extractor only needs the normalization that forward direction operation can extract all detection zones
Target feature vector afterwards does not extract framed feature only once, and simplifies similarity calculation, is algorithm real-time
Provide guarantee.Algorithm creates different goal-based assessment devices to different targets, and the input of each evaluator is characterized extractor
The feature vector of extraction.Goal-based assessment device collects positive sample using k-mean mode, maintains the diversity and of positive sample posture
Weighing apparatus property, the performance for promoting network have biggish help.Goal-based assessment device is updated, only when posture varies widely
When just update current goal evaluator, reduce update times, the real-time of further boosting algorithm；Also network is allowed to fit in time simultaneously
The variation for answering target promotes the performance of network.Pass through roi-pooling layers general of feature extractor combination and normalize
Forward direction operation of layer extracts the feature vector after all detection block normalization, simplifies a large amount of calculate.Feature extractor is mentioned
Input of the feature vector taken as goal-based assessment device, without extracting feature again.Difference is collected by way of with k-mean
The positive sample of posture updates goal-based assessment device, maintains the diversity and harmony of sample posture, for promoting the performance of classifier
There is biggish promotion.In object tracking process, goal-based assessment device only has large change Shi Caigeng new model in posture, significantly
Reduce model modification number；And so that network is adapted to the variation of target in time, promote network performance.
As shown in figure 4, other embodiments of the invention provide a kind of method for tracking target, include the following steps:
S110, tracking target frame is obtained from initial frame image, according to the tracking target frame initialized target tracker
With goal-based assessment device, the feature vector in the tracking target frame is extracted as template characteristic vector；Using initial frame image and
Tracking target frame is trained initialized target tracker to target tracker, is then randomly formed frame in initial frame image,
With the friendship with tracking target frame and than choosing positive sample and negative sample training objective evaluator, trained input is to utilize feature
The feature vector that extractor forms framed calculating.
It should be noted that step S110 can also be realized on another computing device, in realizing that the present invention is implemented
Tracking, original state can be to have trained tracker and goal-based assessment device, realize method training of the invention
The step of be not required；In order to enable entire method and step operation is smooth, corresponding input relationship is embodied, is said herein
It is bright.
S120, subsequent frame image is obtained, the current location rectangle of target is obtained using the target tracker, and to current
Position rectangular extraction feature vector judges the position score of current location rectangle using goal-based assessment device；
S130, the position according to the target in previous frame image and size, are extended to form mesh in current frame image
Mark detection zone, sliding forms frame to be detected in the object detection area, calculate the feature vector of the frame to be detected with
Similarity between template characteristic vector, taking out the highest threshold value of score frame to be detected is detection block, utilizes goal-based assessment device
Obtain the destination probability score of the detection block；Such as 5 can be chosen as detection block.When detecting target in previous frame,
Position based on previous frame target is detected in expanded search region in the current frame, in detection process, is slided by stepping
Mode extracts feature vector to multiple frames to be detected of formation, calculates corresponding probability score using evaluator.
S140, when judging in the position score and the destination probability score that soprano is greater than threshold value, assert tracking at
Function, and using the image in the corresponding detection block of top score or current location rectangle as target.
The extraction process of features described above is extracted using VGG16 model, obtains target frame, detection block to operation using preceding
Feature vector, being capable of quick obtaining feature vector.
Further, the step S140 further includes judging highest in the position score and the destination probability score
When person is less than threshold value, tracking failure is assert；In the step S130, the position according to the target in previous frame image and big
Small, being extended in current frame image and forming object detection area includes: when judging the failure of previous frame image trace, to set institute
State the whole region that object detection area is current frame image.Cope with the case where target is lost.
Further, as shown in figure 5, can specifically be further comprised the steps of: with update module
When tracking successfully in S150, judgment step S140, according to the target feature vector and current mould in current frame image
Plate features vector updates the template characteristic vector, repeats step S120 to step S150 until image terminates.
Specifically, the step of update template characteristic vector includes: to be updated according to the following formula,
FeatTmp=featTmp × (1-alpha)+featCur × alpha, wherein featTmp is template characteristic vector, featCur
For present frame target feature vector, alpha is learning rate.
Further, it as shown in fig. 6, model modification can be carried out, specifically further comprises the steps of:
When tracking successfully in S160, judgment step S140, the feature vector for calculating current goal is commented with for training objective
The target is added to by the similarity for estimating the feature vector of the sample in the sample pool of device when judging that similarity is less than threshold value
In the sample pool, updated goal-based assessment device is formed using the training of updated sample pool, is commented using updated target
Device is estimated as the goal-based assessment device, repeats step S120 to step S160 until image terminates.It is only enough in similarity
Model modification is carried out when low (when morphological differences is larger), the number of model modification can be reduced, while guaranteeing tracking quality；Into
One step, the target obtained in time domain can also be clustered to (such as k-means) by the target after cluster and carry out similarity
Judgement, carries out model modification, in this way, further ensuring the frequency of model modification when less than threshold value.The ratio of positive negative sample can
To control in 1:3 or so.
Further, according to the tracking target frame initialized target tracker and goal-based assessment device in the step S110
The step of include: that regression training is carried out according to the target following frame and initial frame image, obtain the target tracker.
Further, according to the tracking target frame initialized target tracker and goal-based assessment device in the step S110
The step of include: to generate to calculate with machine frame described with machine frame and the friendship of the tracking target frame and ratio in initial frame image, sentence
Break the friendship and when than being greater than first threshold (such as 0.7), will it is corresponding with machine frame as positive sample, judge described hand over and than small
When second threshold (0.3), by corresponding with machine frame negative sample the most, formed using the positive sample and negative sample training
The goal-based assessment device.
Target tracking algorism through the embodiment of the present invention, can efficiently carry out target following, and feature extractor utilizes
Forward direction operation, batch processing frame form feature vector, carry out assessment probability score to feature vector using evaluator later, count
It calculates quick.Meanwhile when more new template and evaluator model, training is re-started using the biggish sample of attitudes vibration, so that dynamic
State better performances, meanwhile, reduce the frequency of update.
Based on above-mentioned method for tracking target, as shown in fig. 7, the embodiment of the invention provides a kind of target trackers
100, comprising:
Initialization module 110, it is initial according to the tracking target frame for obtaining tracking target frame from initial frame image
Change target tracker and goal-based assessment device, extracts the feature vector in the tracking target frame as template characteristic vector；
Position score obtains module 120, for obtaining subsequent frame image, obtains working as target using the target tracker
Front position rectangle, and to current location rectangular extraction feature vector, the position of current location rectangle is judged using goal-based assessment device
Score；
Destination probability obtains module 130, according to the position of the target in previous frame image and size, in current frame image
It is extended to form object detection area, sliding forms frame to be detected in the object detection area, calculates described to be detected
Similarity between the feature vector and template characteristic vector of frame, taking out the highest threshold value of score frame to be detected is detection block,
The destination probability score of the detection block is obtained using goal-based assessment device；
Target Acquisition module 140, for judging that soprano is greater than threshold in the position score and the destination probability score
When value, identification is tracked successfully, and using the image in the corresponding detection block of top score or current location rectangle as target.
Further, the Target Acquisition module is also used to, and is judged in the position score and the destination probability score
When soprano is less than threshold value, tracking failure is assert；The destination probability obtains module and is also used to, described according in previous frame image
Target position and size, being extended in current frame image and forming object detection area includes: to judge previous frame image
When tracking failure, the object detection area is set as the whole region of current frame image.
Further, it further comprises the steps of:
Template renewal module 150, for when target following success when, according in current frame image target feature vector and
Current template feature vector updates the template characteristic vector, so that repeatable position score obtains module, destination probability obtains mould
Block, Target Acquisition module and template renewal module rerun until image procossing terminates.
Specifically, the template renewal module is used for, and is updated according to the following formula, and featTmp=featTmp ×
(1-alpha)+featCur × alpha, wherein featTmp be template characteristic vector, featCur be present frame target signature to
Amount, alpha is learning rate.
It further, further include evaluator update module 160, for calculating current goal when target following success
The similarity of feature vector and the feature vector for the sample in the sample pool of training objective evaluator, judges that similarity is less than
When threshold value, the target is added in the sample pool, updated target is formed using the training of updated sample pool and comments
Device is estimated, using updated goal-based assessment device as the goal-based assessment device, so that position score obtains module, destination probability obtains
Modulus block, Target Acquisition module, template renewal module and evaluator update module or position score obtain module, destination probability
Module, Target Acquisition module and evaluator update module is obtained to rerun until image procossing is complete.
The initialization module 110 is specifically also used to, and carries out recurrence instruction according to the target following frame and initial frame image
Practice, obtains the target tracker.
The initialization module is specifically also used to, and is generated in initial frame image with machine frame, is calculated described with machine frame and institute
State the friendship of tracking target frame and ratio, judge the friendship and when than greater than first threshold, using it is corresponding with machine frame as positive sample, sentence
Break it is described friendship and than be less than second threshold when, by corresponding with machine frame negative sample the most, utilize the positive sample and the negative sample
This training forms the goal-based assessment device.
It further, further include image collecting device, image memory device, user interaction means, image collecting device is used for
Image is acquired, image memory device is transferred to, initialization module obtains information from image memory device and user interaction means,
It is initialized, then position obtains sub-module and destination probability and obtains module obtains from subsequent image carries out from image memory device
Reason.
Target tracker through the embodiment of the present invention, can efficiently carry out target following, and feature extractor utilizes
Forward direction operation, batch processing frame form feature vector, carry out assessment probability score to feature vector using evaluator later, count
It calculates quick.Meanwhile when more new template and evaluator model, training is re-started using the biggish sample of attitudes vibration, so that dynamic
State better performances, meanwhile, reduce the frequency of update.
It has been described in detail in the specific execution step of above-mentioned modules step corresponding in method for tracking target,
It does not do and excessively repeats herein.
Below with reference to Fig. 8, it illustrates the computer systems 800 for the control equipment for being suitable for being used to realize the embodiment of the present application
Structural schematic diagram.Control equipment shown in Fig. 8 is only an example, function to the embodiment of the present application and should not use model
Shroud carrys out any restrictions.
As shown in figure 8, computer system 800 includes central processing unit (CPU) 801, it can be read-only according to being stored in
Program in memory (ROM) 802 or be loaded into the program in random access storage device (RAM) 803 from storage section 808 and
Execute various movements appropriate and processing.In RAM 803, also it is stored with system 800 and operates required various programs and data.
CPU 801, ROM 802 and RAM 803 are connected with each other by bus 804.Input/output (I/O) interface 805 is also connected to always
Line 804.
I/O interface 805 is connected to lower component: the importation 806 including keyboard, mouse etc.；It is penetrated including such as cathode
The output par, c 807 of spool (CRT), liquid crystal display (LCD) etc. and loudspeaker etc.；Storage section 808 including hard disk etc.；
And the communications portion 809 of the network interface card including LAN card, modem etc..Communications portion 809 via such as because
The network of spy's net executes communication process.Driver 810 is also connected to I/O interface 805 as needed.Detachable media 811, such as
Disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on as needed on driver 810, in order to read from thereon
Computer program be mounted into storage section 808 as needed.
Particularly, in accordance with an embodiment of the present disclosure, it may be implemented as computer above with reference to the process of flow chart description
Software program.For example, embodiment of the disclosure includes a kind of computer program product comprising be carried on computer-readable medium
On computer program, which includes the program code for method shown in execution flow chart.In such reality
It applies in example, which can be downloaded and installed from network by communications portion 809, and/or from detachable media
811 are mounted.When the computer program is executed by central processing unit (CPU) 801, limited in execution the present processes
Above-mentioned function.
It should be noted that computer-readable medium described herein can be computer-readable signal media or meter
Calculation machine readable storage medium storing program for executing either the two any combination.Computer readable storage medium for example can be --- but not
Be limited to --- electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor system, device or device, or any above combination.Meter
The more specific example of calculation machine readable storage medium storing program for executing can include but is not limited to: have the electrical connection, just of one or more conducting wires
Taking formula computer disk, hard disk, random access storage device (RAM), read-only memory (ROM), erasable type may be programmed read-only storage
Device (EPROM or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), light storage device, magnetic memory device,
Or above-mentioned any appropriate combination.In this application, computer readable storage medium can be it is any include or storage journey
The tangible medium of sequence, the program can be commanded execution system, device or device use or in connection.And at this
In application, computer-readable signal media may include in a base band or as carrier wave a part propagate data-signal,
Wherein carry computer-readable program code.The data-signal of this propagation can take various forms, including but unlimited
In electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also be that computer can
Any computer-readable medium other than storage medium is read, which can send, propagates or transmit and be used for
By the use of instruction execution system, device or device or program in connection.Include on computer-readable medium
Program code can transmit with any suitable medium, including but not limited to: wireless, electric wire, optical cable, RF etc. are above-mentioned
Any appropriate combination.
The calculating of the operation for executing the application can be write with one or more programming languages or combinations thereof
Machine program code, described program design language include object-oriented programming language-such as Python, Java,
Smalltalk, C++ further include conventional procedural programming language-such as " C " language or similar program design language
Speech.Program code can be executed fully on the user computer, partly be executed on the user computer, as an independence
Software package execute, part on the user computer part execute on the remote computer or completely in remote computer or
It is executed on server.In situations involving remote computers, remote computer can pass through the network of any kind --- packet
It includes local area network (LAN) or wide area network (WAN)-is connected to subscriber computer, or, it may be connected to outer computer (such as benefit
It is connected with ISP by internet).
Flow chart and block diagram in attached drawing are illustrated according to the system of the various embodiments of the application, method and computer journey
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part of one module, program segment or code of table, a part of the module, program segment or code include one or more use
The executable instruction of the logic function as defined in realizing.It should also be noted that in some implementations as replacements, being marked in box
The function of note can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are actually
It can be basically executed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.Also it to infuse
Meaning, the combination of each box in block diagram and or flow chart and the box in block diagram and or flow chart can be with holding
The dedicated hardware based system of functions or operations as defined in row is realized, or can use specialized hardware and computer instruction
Combination realize.
Being described in unit involved in the embodiment of the present application can be realized by way of software, can also be by hard
The mode of part is realized.Described unit also can be set in the processor, for example, can be described as: a kind of processor packet
Include acquiring unit, cutting unit, determination unit and selecting unit.Wherein, the title of these units not structure under certain conditions
The restriction of the pairs of unit itself, for example, acquiring unit is also described as " obtaining the unit to be processed for drawing this image ".
As on the other hand, present invention also provides a kind of computer-readable medium, which be can be
Included in electronic equipment described in above-described embodiment；It is also possible to individualism, and without in the supplying electronic equipment.
Above-mentioned computer-readable medium carries one or more program, when said one or multiple programs are held by the electronic equipment
When row, so that the electronic equipment: obtaining picture frame, the current location rectangle of target is obtained using the target tracker, and right
Current location rectangular extraction feature vector judges the position score of current location rectangle using goal-based assessment device；According to previous frame
The position of target in image and size, are extended to form object detection area in current frame image, examine in the target
It surveys sliding in region and forms frame to be detected, calculate similar between the feature vector and template characteristic vector of the frame to be detected
Degree, taking out the highest threshold value of score frame to be detected is detection block, and the target for obtaining the detection block using goal-based assessment device is general
Rate score；When judging that soprano is greater than threshold value in the position score and the destination probability score, identification is tracked successfully, and will
Image in the corresponding detection block of top score or current location rectangle is as target.
Above description is only the preferred embodiment of the application and the explanation to institute's application technology principle.Those skilled in the art
Member is it should be appreciated that invention scope involved in the application, however it is not limited to technology made of the specific combination of above-mentioned technical characteristic
Scheme, while should also cover in the case where not departing from foregoing invention design, it is carried out by above-mentioned technical characteristic or its equivalent feature
Any combination and the other technical solutions formed.Such as features described above has similar function with (but being not limited to) disclosed herein
Can technical characteristic replaced mutually and the technical solution that is formed.
Claims (10)
1. a kind of method for tracking target, which comprises the steps of:
S120, picture frame is obtained, the current location rectangle of target is obtained using target tracker, and to current location rectangular extraction
Feature vector judges the position score of current location rectangle using goal-based assessment device；
S130, the position according to the target in previous frame image and size are extended to form target inspection in current frame image
Region is surveyed, sliding forms frame to be detected in the object detection area, calculates the feature vector and template of the frame to be detected
Similarity between feature vector, taking out the highest threshold value of score frame to be detected is detection block, is obtained using goal-based assessment device
The destination probability score of the detection block；
S140, when judging in the position score and the destination probability score that soprano is greater than threshold value, identification tracks successfully, and
Using the image in the corresponding detection block of top score or current location rectangle as target.
2. method for tracking target according to claim 1, which is characterized in that the step S140 further includes, described in judgement
When soprano is less than threshold value in position score and the destination probability score, tracking failure is assert；It is described in the step S130
According to the position of the target in previous frame image and size, it is extended to form object detection area packet in current frame image
It includes: when judging the failure of previous frame image trace, setting the object detection area as the whole region of current frame image.
3. method for tracking target according to claim 1 or 2, which is characterized in that further comprise the steps of:
When tracking successfully in S150, judgment step S140, according in current frame image target feature vector and current template it is special
It levies vector and updates the template characteristic vector, repeat step S120 to step S150 until image terminates.
4. method for tracking target according to claim 3, which is characterized in that the step for updating the template characteristic vector
It suddenly include: to be updated according to the following formula, featTmp=featTmp × (1-alpha)+featCur × alpha, wherein
FeatTmp is template characteristic vector, and featCur is present frame target feature vector, and alpha is learning rate.
5. method for tracking target according to claim 1 or 2 or 3, which is characterized in that further comprise the steps of:
When tracking successfully in S160, judgment step S140, the feature vector of current goal is calculated and for training objective evaluator
Sample pool in sample feature vector similarity, when judging that similarity is less than threshold value, the target is added to described
In sample pool, updated goal-based assessment device is formed using the training of updated sample pool, using updated goal-based assessment device
As the goal-based assessment device, step S120 to step S160 is repeated until image terminates.
6. method for tracking target according to claim 1 or 2, which is characterized in that further include: S110, from initial frame image
Middle acquisition tracking target frame extracts the tracking according to the tracking target frame initialized target tracker and goal-based assessment device
Feature vector in target frame is as template characteristic vector；
It according to the step of tracking target frame initialized target tracker and goal-based assessment device include: root in the step S110
Regression training is carried out according to the target following frame and initial frame image, obtains the target tracker.
7. method for tracking target according to claim 6, which is characterized in that further include, according to institute in the step S110
The step of stating tracking target frame initialized target tracker and goal-based assessment device includes: to be generated in initial frame image with machine frame,
It calculates described with machine frame and the friendship of the tracking target frame and ratio, judge the friendship and when than greater than first threshold, it will be corresponding
, by corresponding with machine frame negative sample the most, institute is utilized when judging the friendship and comparing less than second threshold as positive sample with machine frame
It states positive sample and negative sample training forms the goal-based assessment device.
8. a kind of target tracker characterized by comprising
Position score obtains module, and for obtaining picture frame, the current location rectangle of target is obtained using the target tracker,
And to current location rectangular extraction feature vector, the position score of current location rectangle is judged using goal-based assessment device；
Destination probability obtains module and is expanded in current frame image according to the position of the target in previous frame image and size
Spread is at object detection area, and sliding forms frame to be detected in the object detection area, calculates the spy of the frame to be detected
The similarity between vector and template characteristic vector is levied, taking out the highest threshold value of score frame to be detected is detection block, utilizes mesh
Mark evaluator obtains the destination probability score of the detection block；
Target Acquisition module is recognized when for judging that soprano is greater than threshold value in the position score and the destination probability score
It tracks successfully calmly, and using the image in the corresponding detection block of top score or current location rectangle as target.
9. a kind of electronic equipment, comprising:
One or more processors；
Storage device is stored thereon with one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processors are real
The now method as described in any in claim 1-7.
10. a kind of computer-readable medium, is stored thereon with computer program, wherein real when described program is executed by processor
The now method as described in any in claim 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811516817.9A CN109671103A (en) | 2018-12-12 | 2018-12-12 | Method for tracking target and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811516817.9A CN109671103A (en) | 2018-12-12 | 2018-12-12 | Method for tracking target and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109671103A true CN109671103A (en) | 2019-04-23 |
Family
ID=66143780
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811516817.9A Pending CN109671103A (en) | 2018-12-12 | 2018-12-12 | Method for tracking target and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109671103A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110211153A (en) * | 2019-05-28 | 2019-09-06 | 浙江大华技术股份有限公司 | Method for tracking target, target tracker and computer storage medium |
| CN110222686A (en) * | 2019-05-27 | 2019-09-10 | 腾讯科技（深圳）有限公司 | Object detecting method, device, computer equipment and storage medium |
| CN110400329A (en) * | 2019-06-17 | 2019-11-01 | 北京百度网讯科技有限公司 | People stream counting method and its system |
| CN110661977A (en) * | 2019-10-29 | 2020-01-07 | Oppo广东移动通信有限公司 | Subject detection method and apparatus, electronic device, and computer-readable storage medium |
| CN110688930A (en) * | 2019-09-20 | 2020-01-14 | Oppo广东移动通信有限公司 | Face detection method, face detection device, mobile terminal and storage medium |
| CN110910422A (en) * | 2019-11-13 | 2020-03-24 | 北京环境特性研究所 | Target tracking method and device, electronic equipment and readable storage medium |
| CN111242973A (en) * | 2020-01-06 | 2020-06-05 | 上海商汤临港智能科技有限公司 | Target tracking method and device, electronic equipment and storage medium |
| CN111242981A (en) * | 2020-01-21 | 2020-06-05 | 北京捷通华声科技股份有限公司 | Tracking method and device for fixed object and security equipment |
| CN111402291A (en) * | 2020-03-04 | 2020-07-10 | 北京百度网讯科技有限公司 | Method and apparatus for tracking a target |
| CN111832549A (en) * | 2020-06-29 | 2020-10-27 | 深圳市优必选科技股份有限公司 | Data labeling method and device |
| CN111914831A (en) * | 2019-05-10 | 2020-11-10 | 杭州海康威视数字技术股份有限公司 | Target detection method, device and storage medium |
| WO2021036373A1 (en) * | 2019-08-27 | 2021-03-04 | 北京京东尚科信息技术有限公司 | Target tracking method and device, and computer readable storage medium |
| CN112631333A (en) * | 2020-12-25 | 2021-04-09 | 南方电网数字电网研究院有限公司 | Target tracking method and device of unmanned aerial vehicle and image processing chip |
| CN112784926A (en) * | 2021-02-07 | 2021-05-11 | 四川长虹电器股份有限公司 | Gesture interaction method and system |
| CN113486820A (en) * | 2021-07-09 | 2021-10-08 | 厦门理工学院 | Bidirectional target tracking method and system based on efficient template updating and selecting mechanism |
| WO2021227351A1 (en) * | 2020-05-15 | 2021-11-18 | 北京百度网讯科技有限公司 | Target part tracking method and apparatus, electronic device and readable storage medium |
| CN113989694A (en) * | 2021-09-18 | 2022-01-28 | 北京远度互联科技有限公司 | Target tracking method and device, electronic equipment and storage medium |
| CN115375929A (en) * | 2022-10-25 | 2022-11-22 | 杭州华橙软件技术有限公司 | Target template set updating method and device and computer readable storage medium |
| CN116993785A (en) * | 2023-08-31 | 2023-11-03 | 东之乔科技有限公司 | Target object visual tracking method and device, electronic equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107066990A (en) * | 2017-05-04 | 2017-08-18 | 厦门美图之家科技有限公司 | A kind of method for tracking target and mobile device |
| CN107886048A (en) * | 2017-10-13 | 2018-04-06 | 西安天和防务技术股份有限公司 | Method for tracking target and system, storage medium and electric terminal |
| WO2018121286A1 (en) * | 2016-12-30 | 2018-07-05 | 纳恩博（北京）科技有限公司 | Target tracking method and device |
| CN108765452A (en) * | 2018-05-11 | 2018-11-06 | 西安天和防务技术股份有限公司 | A kind of detection of mobile target in complex background and tracking |
-
2018
- 2018-12-12 CN CN201811516817.9A patent/CN109671103A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018121286A1 (en) * | 2016-12-30 | 2018-07-05 | 纳恩博（北京）科技有限公司 | Target tracking method and device |
| CN107066990A (en) * | 2017-05-04 | 2017-08-18 | 厦门美图之家科技有限公司 | A kind of method for tracking target and mobile device |
| CN107886048A (en) * | 2017-10-13 | 2018-04-06 | 西安天和防务技术股份有限公司 | Method for tracking target and system, storage medium and electric terminal |
| CN108765452A (en) * | 2018-05-11 | 2018-11-06 | 西安天和防务技术股份有限公司 | A kind of detection of mobile target in complex background and tracking |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111914831A (en) * | 2019-05-10 | 2020-11-10 | 杭州海康威视数字技术股份有限公司 | Target detection method, device and storage medium |
| CN110222686A (en) * | 2019-05-27 | 2019-09-10 | 腾讯科技（深圳）有限公司 | Object detecting method, device, computer equipment and storage medium |
| CN110211153A (en) * | 2019-05-28 | 2019-09-06 | 浙江大华技术股份有限公司 | Method for tracking target, target tracker and computer storage medium |
| CN110400329A (en) * | 2019-06-17 | 2019-11-01 | 北京百度网讯科技有限公司 | People stream counting method and its system |
| CN110400329B (en) * | 2019-06-17 | 2022-04-05 | 北京百度网讯科技有限公司 | People flow counting method and system |
| WO2021036373A1 (en) * | 2019-08-27 | 2021-03-04 | 北京京东尚科信息技术有限公司 | Target tracking method and device, and computer readable storage medium |
| CN110688930A (en) * | 2019-09-20 | 2020-01-14 | Oppo广东移动通信有限公司 | Face detection method, face detection device, mobile terminal and storage medium |
| CN110661977B (en) * | 2019-10-29 | 2021-08-03 | Oppo广东移动通信有限公司 | Subject detection method and apparatus, electronic device, and computer-readable storage medium |
| WO2021082883A1 (en) * | 2019-10-29 | 2021-05-06 | Oppo广东移动通信有限公司 | Main body detection method and apparatus, and electronic device and computer readable storage medium |
| CN110661977A (en) * | 2019-10-29 | 2020-01-07 | Oppo广东移动通信有限公司 | Subject detection method and apparatus, electronic device, and computer-readable storage medium |
| CN110910422A (en) * | 2019-11-13 | 2020-03-24 | 北京环境特性研究所 | Target tracking method and device, electronic equipment and readable storage medium |
| CN111242973A (en) * | 2020-01-06 | 2020-06-05 | 上海商汤临港智能科技有限公司 | Target tracking method and device, electronic equipment and storage medium |
| CN111242981A (en) * | 2020-01-21 | 2020-06-05 | 北京捷通华声科技股份有限公司 | Tracking method and device for fixed object and security equipment |
| CN111402291A (en) * | 2020-03-04 | 2020-07-10 | 北京百度网讯科技有限公司 | Method and apparatus for tracking a target |
| CN111402291B (en) * | 2020-03-04 | 2023-08-29 | 阿波罗智联(北京)科技有限公司 | Method and apparatus for tracking a target |
| WO2021227351A1 (en) * | 2020-05-15 | 2021-11-18 | 北京百度网讯科技有限公司 | Target part tracking method and apparatus, electronic device and readable storage medium |
| CN111832549A (en) * | 2020-06-29 | 2020-10-27 | 深圳市优必选科技股份有限公司 | Data labeling method and device |
| CN111832549B (en) * | 2020-06-29 | 2024-04-23 | 深圳市优必选科技股份有限公司 | Data labeling method and device |
| CN112631333A (en) * | 2020-12-25 | 2021-04-09 | 南方电网数字电网研究院有限公司 | Target tracking method and device of unmanned aerial vehicle and image processing chip |
| CN112631333B (en) * | 2020-12-25 | 2024-04-12 | 南方电网数字电网研究院有限公司 | Target tracking method and device of unmanned aerial vehicle and image processing chip |
| CN112784926A (en) * | 2021-02-07 | 2021-05-11 | 四川长虹电器股份有限公司 | Gesture interaction method and system |
| CN113486820B (en) * | 2021-07-09 | 2023-06-06 | 厦门理工学院 | Bidirectional target tracking method and system based on efficient template updating and selecting mechanism |
| CN113486820A (en) * | 2021-07-09 | 2021-10-08 | 厦门理工学院 | Bidirectional target tracking method and system based on efficient template updating and selecting mechanism |
| CN113989694B (en) * | 2021-09-18 | 2022-10-14 | 北京远度互联科技有限公司 | Target tracking method and device, electronic equipment and storage medium |
| CN113989694A (en) * | 2021-09-18 | 2022-01-28 | 北京远度互联科技有限公司 | Target tracking method and device, electronic equipment and storage medium |
| CN115375929A (en) * | 2022-10-25 | 2022-11-22 | 杭州华橙软件技术有限公司 | Target template set updating method and device and computer readable storage medium |
| CN115375929B (en) * | 2022-10-25 | 2023-02-07 | 杭州华橙软件技术有限公司 | Target template set updating method and device and computer readable storage medium |
| CN116993785A (en) * | 2023-08-31 | 2023-11-03 | 东之乔科技有限公司 | Target object visual tracking method and device, electronic equipment and storage medium |
| CN116993785B (en) * | 2023-08-31 | 2024-02-02 | 东之乔科技有限公司 | Target object visual tracking method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109671103A (en) | Method for tracking target and device | |
| CN109099903B (en) | Method and apparatus for generating navigation routine | |
| CN108171207A (en) | Face identification method and device based on video sequence | |
| CN111598164B (en) | Method, device, electronic equipment and storage medium for identifying attribute of target object | |
| CN110506276A (en) | The efficient image analysis of use environment sensing data | |
| CN108960090A (en) | Method of video image processing and device, computer-readable medium and electronic equipment | |
| CN108304835A (en) | character detecting method and device | |
| CN108229335A (en) | It is associated with face identification method and device, electronic equipment, storage medium, program | |
| CN109584276A (en) | Critical point detection method, apparatus, equipment and readable medium | |
| CN106874826A (en) | Face key point-tracking method and device | |
| CN108280477A (en) | Method and apparatus for clustering image | |
| CN108229418B (en) | Human body key point detection method and apparatus, electronic device, storage medium, and program | |
| CN110222686B (en) | Object detection method, object detection device, computer equipment and storage medium | |
| CN109644255A (en) | Mark includes the method and apparatus of the video flowing of a framing | |
| CN108229456A (en) | Method for tracking target and device, electronic equipment, computer storage media | |
| CN110443824A (en) | Method and apparatus for generating information | |
| CN108288051A (en) | Pedestrian identification model training method and device, electronic equipment and storage medium again | |
| CN107918767B (en) | Object detection method, device, electronic equipment and computer-readable medium | |
| CN109325456A (en) | Target identification method, device, target identification equipment and storage medium | |
| CN111914812A (en) | Image processing model training method, device, equipment and storage medium | |
| CN108509921A (en) | Method and apparatus for generating information | |
| CN112699832B (en) | Target detection method, device, equipment and storage medium | |
| CN108985190A (en) | Target identification method and device, electronic equipment, storage medium, program product | |
| CN109345460B (en) | Method and apparatus for rectifying image | |
| CN113111947A (en) | Image processing method, apparatus and computer-readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |