CN109948582A - A kind of retrograde intelligent detecting method of vehicle based on pursuit path analysis - Google Patents
A kind of retrograde intelligent detecting method of vehicle based on pursuit path analysis Download PDFInfo
- Publication number
- CN109948582A CN109948582A CN201910244210.8A CN201910244210A CN109948582A CN 109948582 A CN109948582 A CN 109948582A CN 201910244210 A CN201910244210 A CN 201910244210A CN 109948582 A CN109948582 A CN 109948582A
- Authority
- CN
- China
- Prior art keywords
- target vehicle
- vehicle
- target
- frame
- tracking
- 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.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of retrograde intelligent detecting methods of the vehicle based on pursuit path analysis, it detects the target vehicle in video image using first-level class device and secondary classifier, extract target vehicle region, and nuclear phase is distributed for target vehicle and closes filter tracker, each target vehicle matches a nuclear phase and closes filter tracker, the tracing area that filter tracker obtains target vehicle is closed using nuclear phase, and then the motion profile direction of growth of target vehicle is obtained according to tracing area, it is compared with the retrograde direction of initial markers, if they are the same, target vehicle occurs to drive in the wrong direction;If it is different, target vehicle does not drive in the wrong direction.The present invention realizes the real-time detection driven in the wrong direction to vehicle by this way, solves the problems of artificial detection;Tracking while improving the reliability of vehicle detection identification using cascade classifier simultaneously, and realize more vehicle targets.
Description
Technical field
The invention belongs to technical field of computer vision, and in particular to a kind of retrograde intelligence of vehicle based on pursuit path analysis
It can detection method.
Background technique
With the continuous development of society, road vehicle is more and more, and traffic safety problem becomes increasingly conspicuous, it is annual because
Traffic casualties are startling, and because traffic accident caused by driving in the wrong direction often causes great harm on road.
Attention now for road safety is continuously improved, and a large amount of camera is mounted on road for monitoring, for these magnanimity
Monitor video data, mostly in the past was to carry out judging whether to occur by manually abnormal, can not only consume a large amount of manpowers in this way
And it unremitting can not work within 24 hours.With the development of intelligent Video Surveillance Technology, some basic road traffics
Anomalous event may be implemented to detect automatically.
The current retrograde detection method of existing vehicle is mostly to extract movement mesh using the methods of optical flow method, background modeling
Mark, then realizes the tracking of vehicle, and then judge whether vehicle drives in the wrong direction using track of vehicle information by similarity mode.This
Kind method is realized simply, but if it is a problem that velocity to moving target, which slowly detects, for optical flow method, background modeling is more
It is susceptible to the influence of environment.At present in the research method of automobile detecting following, have for the traditional method of detection part
Based on priori knowledge, vehicle is detected using priori knowledges such as vehicle bottom shadow, tail-lights;Or use more advanced Haar_
Like feature, HOG feature are trained classifier to detect vehicle, but are difficult to using the single classifier that single feature trains
It is applicable in traffic scene complicated and changeable;Alternatively, using the current popular detection method based on convolutional neural networks, although
Significant effect improves, but because its is computationally intensive, is difficult to the real time execution in common hardware equipment terminal.For vehicle target
Tracking, previous method be mostly using Kalman filtering, particle filter etc., these methods single goal can only generally be carried out with
Track, and once tracking error cannot be corrected in time, effect is unreliable.
Therefore, a kind of efficient, the retrograde detection method of reliable vehicle and existing vehicle detection are lacked for the prior art
The classification reliability of single classifier is to be improved in technology and existing wireless vehicle tracking is to carry out track side to single goal
The defect problem in face is badly in need of a kind of side based on the retrograde detection of the vehicle realized in reliable moving vehicles detection and tracking technical foundation
Method.
Summary of the invention
The present invention provides a kind of retrograde intelligent detecting method of the vehicle based on pursuit path analysis, may be implemented on highway
The quick and precisely identification of retrograde vehicle, while during vehicle detection, it is carried out using the classifier of two off-line trainings secondary
Identification improves the reliability of vehicle detection result;The tracking to more vehicle targets is realized simultaneously, expands its application range.
On the one hand, the retrograde intelligent detecting method of a kind of vehicle based on pursuit path analysis provided by the invention, including such as
Lower step:
S1: retrograde detection zone and retrograde bearing mark on camera shooting area image are obtained and is obtained in real time and is taken the photograph
As the video data of head;
Wherein, detection zone of driving in the wrong direction is the quadrilateral area comprising vehicle driving lane, and retrograde approach is labeled as having direction
Line segment;
S2: first-level class device and secondary classifier based on off-line training are mentioned from video data when in previous frame image
Take the target vehicle region in retrograde detection zone;
Wherein, Classification and Identification is carried out to current frame image using first-level class device and obtains target vehicle candidate region, then benefit
Classification and Identification is carried out to the target vehicle candidate region with secondary classifier and obtains the target vehicle region on current frame image;
Wherein, the input data of the first-level class device and the secondary classifier off-line training process is positive sample respectively
Image, negative sample image, include vehicle in the positive sample image, and the negative sample image does not include vehicle;
S3: correlation filtering tracker is distributed based on the target vehicle region that step S2 is extracted, each nuclear phase closes filter tracking
Device obtains corresponding mesh based on other frame image trace target vehicles in target vehicle region in current frame image and video data
Mark the tracking information of vehicle;
One nuclear phase closes filter tracker and tracks a target vehicle, and current all nuclear phases close filter tracker and corresponding mesh
It marks vehicle and constitutes current tracking queue;
Wherein, other frame images in video data are successively handled according to step S2 and step S3, and if depositing in treatment process
In the target vehicle identified not in current tracking queue, the target vehicle is fresh target vehicle, and distributes a nuclear phase
Filter tracker is closed for tracking the fresh target vehicle;
S4: whether tracking information and retrograde bearing mark based on target vehicle identify target vehicle in retrograde detection zone
Domain occurs to drive in the wrong direction;
Wherein, tracking information includes target detection frame position of the target vehicle in sequential frame image, the target detection
Frame indicates target vehicle in the tracing area of each frame image;The position of target detection frame based on target vehicle in sequential frame image
The motion profile direction of growth for obtaining same position on target detection frame is set, and judges the motion profile growing method and drives in the wrong direction
Whether direction is identical, and if they are the same, corresponding target vehicle occurs to drive in the wrong direction;If not identical, there is no driving in the wrong direction.
The present invention is by the motion profile growing method of tracking target vehicle, and by itself and retrograde approach mark initially set
Note is compared, and then obtains the conclusion whether vehicle drives in the wrong direction, and is provided a kind of be automatically brought into operation and is identified the hand whether vehicle drives in the wrong direction
Section;Meanwhile the present invention is that the classifier based on two ranks carries out target vehicle identification first, compared to the knowledge of single classifier
Not as a result, the reliability for the recognition result that mode provided by the invention obtains is higher.In addition, which provided by the invention with
Multiple target tracking is realized during track, each vehicle target distributes a tracker, is added and coincide with practical application.
It further preferably, further include that the tracking information based on target vehicle identifies target vehicle if vehicle does not occur to drive in the wrong direction
Driving status, process is as follows:
Firstly, same on the position acquisition target detection frame of the target detection frame in continuous N1 frame image based on target vehicle
The motion profile growth length of one position;Then, judge whether the motion profile growth length is greater than preset first threshold value L1;
If more than preset first threshold value L1, the target vehicle normally travel;
Otherwise, same on the position acquisition target detection frame of the target detection frame in continuous N2 frame image based on target vehicle
The new motion profile growth length of one position, and judge whether the new motion profile growth length is less than default second threshold
L2;
Wherein, N2 is greater than N1, and L2 is less than L1;
If being less than default second threshold L2, judge that current same period same driving direction whether there is other target carriages
?;
If there are other target vehicles, other target vehicles of same driving direction of same period are obtained in continuous N1 frame
Motion profile growth length;
Wherein, if motion profile growth length of other target vehicles in continuous N1 frame is greater than preset first threshold value L1,
The target vehicle is in abnormality;If motion profile growth length of other target vehicles in continuous N1 frame is respectively less than pre-
If second threshold L2, the target vehicle is in blocked state;If motion profile of other targets in continuous N1 frame grows length
Degree is less than or equal to preset first threshold value L1 and is greater than or equal to second threshold L2, and the target vehicle is in state undetermined;
If the target vehicle is in abnormality without other target vehicles;
If the new motion profile growth length is not less than default second threshold L2, the target vehicle is in shape undetermined
State.
The present invention can also identify the state of target vehicle using motion profile growth length, check abnormal feelings in time
Condition, to regulate and control in time, to releive or being provided some basis for close to traffic problems.
Further preferably, the value range difference of the corresponding preset first threshold value L1 of target vehicle, default second threshold L2
Are as follows: L1>2/3h, L2<1/3h;
Wherein, h is pixel value shared by the height of target detection frame when tracker captures the target vehicle for the first time.
From above-mentioned statement it is found that preset first threshold value L1, default second threshold L2 are the target detections according to target vehicle
It is fixed that frame comes, then the first threshold of the motion profile length for the target vehicle being that is each tracked, second threshold are
It is different, this is because type of vehicle difference leads to target detection frame size and different.But it is preset first threshold value L1, default
The obtaining value method of second threshold L2 is consistent.
Wherein, N1 frame and N2 frame value are empirical value, related with the acquisition frame rate of actual video, and for example camera is adopted
Collection frame per second is bigger, and the value of 1 frame and N2 frame is bigger.100 meters of effective detecting distance or so of general monitoring camera, by high speed
Speed 120km/h, vehicle time of occurrence 3 seconds or so, it is assumed that algorithm effectively identifies that vehicle time only has 2s, and video acquisition frame rate is
It is m, N2 2m that m fps (m frame per second), N1, which can select value,.
Further preferably, the position acquisition mesh of the target detection frame in step S4 based on target vehicle in sequential frame image
Mark the motion profile direction of growth and length of the following center position of detection block;
Wherein, at interval of the coordinate value of the following center position of n1 frame recording target detection frame.
At interval of the coordinate value of the following center position of n1 frame recording target detection frame, and then concatenates and obtain motion profile life
Length direction and length.Wherein, at interval of n1 frame recording coordinate value, rather than the corresponding coordinate value of every frame image, Ke Yifang are recorded
Data redundancy is set, space waste is reduced while the characterization to target trajectory will not be lost.N1 is more than or equal to 2
Positive integer, for example n1 is 5.
Further preferably, further include following steps when handling next frame image:
Firstly, tracing area where the target vehicle in tracking queue current in next frame image is carried out mask;
Wherein, filter tracker is closed using the nuclear phase of target vehicle in current tracking queue obtain target vehicle in next frame
Tracing area in image;
Then, according to the next frame image after step S2 and step S3 processing mask.
The mode of mask is that the pixel value for the pixel for corresponding to tracing area is disposed as 255 in the present invention.To protect
When card carries out target vehicle region detection to next frame image, the target vehicle recognized is fresh target vehicle.Pass through mask
Mode can make corresponding tracing area be determined as negative sample (as non-targeted vehicle in first-level class device, secondary classifier
Region), accelerate detection speed.
It further preferably, further include prison when closing the tracking information of the corresponding target vehicle of filter tracker acquisition using nuclear phase
Superintend and direct whether target vehicle is disappearing within sweep of the eye or reaching the boundary position of retrograde detection zone;
Wherein, if disappearing within sweep of the eye or target vehicle reaches the boundary position of retrograde detection zone, currently with
The nuclear phase filter tracker is closed in track queue to delete;
Wherein, by nuclear phase close filter tracker can predict target vehicle in the tracing area of next frame, and visual field model
It encloses and the position for detection zone of driving in the wrong direction is determining, therefore whether can identify in field range or still also in detection of driving in the wrong direction
In region.
It further preferably, further include pair when closing the tracking information of the corresponding target vehicle of filter tracker acquisition using nuclear phase
Nuclear phase closes filter tracker tracking and carries out punishment score;
Wherein, filter tracker is closed based on nuclear phase and obtains tracing area of the target vehicle in next frame image, and utilized
Whether the vehicle classification device recognition and tracking region of preparatory off-line training includes vehicle;If do not include vehicle, depending on core correlation filtering with
The tracking error of track device, carries out punishment score to it;
Wherein, include vehicle when Probability p is more than or equal to 0.5, when Probability p is less than 0.5, do not include vehicle;Vehicle point
Class device obtain be vehicle probability be p when, then score be 1-p;
Wherein, if the accumulation score that nuclear phase closes filter tracker reaches default S points, by the core in current tracking queue
Correlation filtering tracker is deleted;
The input data of the vehicle classification device training process is positive sample image and negative sample image.
Nuclear phase close filter tracker can have always tracking target (such as in n+30 frame tracker correct mesh
Mark is lost, but tracker also will continue to work, looked in the n-th+31 frame it is most similar with the n-th+30 frame institute tracing area, to the greatest extent
Managing this has been one wrong).Here it is nuclear phases to close the target loss that filter tracker often occurs during tracking.This hair
It is bright to use a vehicle classification device, whether there is tracking target loss to identify.Simultaneously it is also contemplated that actual complex traffic
Classifier can not reach accuracy rate absolutely in scene, so being not the judgement tracking of vehicle classification device without being vehicle
It is abandoned from tracker immediately, but has given punishment score, when punishment reaches certain threshold value S just by it from tracker
It abandons.
Wherein, vehicle classification device is two classifiers, and obtained result is a probability value p in fact, and size is in 0-1.
The only just triggering score when vehicle classification device is not thought being vehicle.S is empirical value in the present invention, is for example taken in the embodiment of the present invention
Value is 2.Each nuclear phase closes filter tracker and only focuses on a target vehicle in the present invention, if reaching penalty threshold S, the nuclear phase
Closing filter tracker will reject from tracking queue, and it is all clear therewith which closes the corresponding classifier of filter tracker, score s
It removes.
Further preferably, the first-level class device is the AdaBoost classifier based on cascade model;The secondary classification
Device is support vector machines;
Wherein, the training process of the first-level class device is instructed based on the Haar_like feature of positive sample and negative sample
Practice;
The training process of the secondary classifier is carried out based on the feature vector that positive sample and the HOG feature of negative sample are constituted
Trained.
Beneficial effect
On the one hand, the present invention by tracking target vehicle motion profile growing method, and by its with it is initially set inverse
Row method label is compared, and then obtains the conclusion whether vehicle drives in the wrong direction, and provides one kind and be automatically brought into operation whether identify vehicle
Retrograde means get rid of the obstacle of personal monitoring.
On the other hand, whether other than it can identify and drive in the wrong direction, state that can also be current to vehicle carries out the present invention
Identification identifies that it is that normally travel is in abnormality or mentions in traffic congestion state to dispatch in time for traffic administration
Data basis is supplied.
In addition, the present invention realizes the process of drive in the wrong direction identification and state identification, require accurately to carry out target vehicle
Recognition and tracking, the present invention carries out target vehicle identification using the classifier of two ranks, compared to the identification of single classifier
As a result, this available higher recognition result of reliability.Meanwhile
Which provided by the invention realizes multiple target tracking during tracking, each vehicle target distribute one with
Track device is added with practical application and coincide.
Detailed description of the invention
Fig. 1 is a kind of flow diagram of the retrograde intelligent detecting method of vehicle based on pursuit path analysis of the invention.
Fig. 2 is automobile detecting following in a kind of retrograde intelligent detecting method of vehicle based on pursuit path analysis of the invention
With the flow diagram of trajectory analysis.
Fig. 3 is that target detection frame moves schematic diagram in vehicle movement trajectory analysis of the invention.
Specific embodiment
Below in conjunction with embodiment, the present invention is described further.
The retrograde intelligent detecting method of a kind of vehicle based on pursuit path analysis provided by the invention, includes the following steps:
S1: retrograde detection zone and retrograde bearing mark on camera shooting area image are obtained and is obtained in real time and is taken the photograph
As the video data of head.
Wherein, due to after monitoring installation in shooting area image lane position and track direction be all uniquely determine and
Fixed, therefore it may only be necessary to carry out an initialization process (the retrograde detection zone of handmarking and retrograde direction) to it.
S2: first-level class device and secondary classifier based on off-line training are mentioned from video data when in previous frame image
Take the target vehicle region in retrograde detection zone.
In the present embodiment, using vehicle detection data set disclosed in the country, in conjunction in the local city of acquisition, super expressway
Traffic image data set cuts out vehicle as positive sample, cuts out the traffic scene not comprising vehicle as negative sample, production
At the positive and negative sample set that training uses, positive and negative sample proportion is about 1:3, positive negative sample is uniformly scaled 24x24.Using just
Negative sample collection off-line training first-level class device and secondary classifier.
Wherein, first-level class device selects the AdaBoost classifier based on cascade model in the present embodiment.Its training process
The Haar_like feature of positive negative sample is calculated to train AdaBoost classifier.It is i.e. special using the edge in Haar_like feature
Sign, linear character quickly calculate sample characteristics integrogram, then pass through the form calculus Haar_like feature of integrogram.In view of with
It is image general characteristics in Haar_like feature, therefore the present invention calculates it acquisition process without being described in detail.The present invention
Sketch the AdaBoost classifier based on cascade model, AdaBoost is as a kind of boosting method, to every when beginning
One training sample assigns an equal initial weight, then trains on training set and carrys out a Weak Classifier, and calculates
This error rates of weak classifiers readjusts the weight of each sample in training set according to obtained Weak Classifier error rate,
Classification error sample is set to obtain higher weight, the correct sample reduction weight of classification.After the weight for having updated each sample,
Start next round training, until reaching the number of iterations set or classifier error rate as zero, then according to these weak typings
Weight be combined into a strong classifier.AdaBoost algorithm is trained the classifier come as in cascade model by the present invention
One strong classifier.Each strong classifier is higher to the determination rate of accuracy of negative sample, when certain level-one strong classifier judge be
When negative sample, the classifier of next stage will not continue to call, and most of in piece image in traffic scene is all negative
Sample, all excessively this mode can just give up a large amount of negative samples at the initial stage of cascade classifier, accelerate detection of classifier speed.
Only previous stage classifier is judged to positive sample just and can be sent to next stage classifier to determine again in cascade classifier simultaneously, guarantees
In the vehicle rough detection stage just reduce the false detection rate of vehicle candidate region.
In summary, the present invention is using the AdaBoost classifier based on cascade model to the video frame images obtained in real time
It carries out vehicle detection and gets target vehicle candidate region.When for example carrying out vehicle detection to current frame image, scanning window will
Image in window is sent into first-level class device, and then identifies whether image in window includes vehicle, if being mesh comprising vehicle
Mark vehicle candidate region;It otherwise is not then to abandon.
Secondary classifier is selected as support vector machines in the present embodiment, calculates HOG feature by positive and negative sample set and obtains HOG
The HOG feature vector being calculated is sent into support vector machines classifier as input vector and instructed by Feature Descriptor
Practice.It wherein, the use of the size of detection window is 56x56, block size 16x16, cell size 8x8, block sliding step 8x8, bin number
Mesh is set as 9.Similarly, since HOG feature is image general characteristics, the present invention calculates it acquisition process without superfluous in detail
It states.The secondary classifier that the present invention constructs in practical applications, is further reflected for the classification results to first-level class device
It is fixed.
For to sum up, the present invention carries out Classification and Identification to current frame image using first-level class device and obtains target vehicle candidate
Region recycles secondary classifier to carry out Classification and Identification to the target vehicle candidate region and obtains the target on current frame image
Vehicle region.
S3: correlation filtering tracker is distributed based on the target vehicle region that step S2 is extracted, each nuclear phase closes filter tracking
Device obtains corresponding mesh based on other frame image trace target vehicles in target vehicle region in current frame image and video data
Mark the tracking information of vehicle.
Firstly, illustratively nuclear phase used in the present invention closes filter tracker (KCF), nuclear phase closes filter tracker
It (KCF) is a monotrack device, the target that some in delineation video flowing determines when tracking process is initial, followed by calculation
Method found in next frame image with an initialized target most similar panel region, the region be considered target present frame with
Track region.The present invention selects KCF algorithm.
Secondly, the present invention constructs a multiple target tracking management module, for tracking to target vehicle.Multiple target with
Track management module is for this case that all more than one target vehicles general in traffic scene, and the fusion designed is multiple
The management module that target following and target timely update.One nuclear phase closes filter tracker and tracks a target vehicle, current institute
There is nuclear phase to close filter tracker and the current tracking queue of corresponding target vehicle composition.Therefore, the same target vehicle is in video flowing
In be that the same core correlation filter is tracked, set the goal the tracking of vehicle in video streaming really using video streaming image
Region.
In order to enable the target vehicle region that other frame image detections go out in video flowing is fresh target vehicle corresponding region, this
When invention next frame image, perform the following operations:
Firstly, tracing area where the target vehicle in tracking queue current in next frame image is carried out mask;Then,
According to the next frame image after step S2 and step S3 processing mask.
Wherein, filter tracker is closed using the nuclear phase of target vehicle in current tracking queue obtain target vehicle in next frame
Then whole pixel values of corresponding tracing area are set to 255 by the tracing area in image.So that according to step S2 and step
When rapid S3 handles next frame image, the target vehicle identified is fresh target vehicle, i.e., not in current tracking queue
In, and then a nuclear phase can be distributed and close filter tracker tracking fresh target vehicle.The present invention by this way can be quick
Identify fresh target region, this is that treated that region is quickly determined the sample that is negative in first-level class device based on mask
This, that is, do not include vehicle.
It should be appreciated that the present invention successively handles other frame images according to step S2 and step S3, simultaneously, nuclear phase closes filter
Wave tracker synchronizes the tracing area for carrying out the corresponding target vehicle of tracking acquisition in other frame images.
In order to enable nuclear phase closes filter tracker there is more tracking effect, the present invention to change to core correlation filter
Into for each target vehicle one On-line monitor device of outfit, whether On-line monitor device returns judges target vehicle within sweep of the eye
It disappears or reaches the boundary position of retrograde detection zone.The for example upper left of target detection frame of the target vehicle in sequential frame image
Angular coordinate or bottom right angular coordinate meet the boundary threshold of setting, then may determine that target vehicle has reached retrograde detection zone
Boundary position, and then it is rejected from tracking queue.
In addition, the present invention goes back one vehicle classification device of off-line training, for the tracking effect to core correlation filter into
Row identification, judgement are whether the target of tracking drifts about.Wherein, vehicle classification device and first-level class device, secondary classifier phase
It together, is for identifying whether image-region includes vehicle, off-line training process is trained also with positive negative sample, tool
Body is the gradient information for utilizing each pixel in positive and negative samples, and gradient direction is divided into 9 bin, counts each bin
Middle pixel number, obtains positive and negative samples gradient direction statistic histogram, which is positive, the feature of negative and positive sample
Vector carries out off-line training using this feature vector.
In the present embodiment, when identifying the tracking effect that a certain nuclear phase closes filter tracker, it is based on core correlation filtering
The tracing area that device is detected in next frame, and the statistic histogram of the tracing area gradient direction is calculated, and then pass through vehicle
Classifier identifies whether tracing area has vehicle, if there is vehicle, it is errorless to be considered as tracking;If being considered as tracking error without vehicle.
Filter tracker is closed for each nuclear phase in the present invention, punishment score is carried out using the above method, every error is primary
When, vehicle classification device obtains when be the probability of vehicle being p, then score is 1-p, if having reached S point, by it from tracking queue
Middle rejecting.
S4: whether tracking information and retrograde bearing mark based on target vehicle identify target vehicle in retrograde detection zone
The driving status for driving in the wrong direction and detecting target vehicle occurs for domain.
In the present embodiment, tracking process of the target vehicle in sequential frame image will obtain target as shown in Figure 3
Detection block.Record the coordinate value of the following center of target detection frame, and then the motion profile growth of available target vehicle
Direction and motion profile growth length.Preferably at interval of the following center of n1 frame recording target detection frame in the present embodiment
Coordinate value, and then space waste is reduced, prevent data redundancy.
The present invention is based on the motion profile directions of growth of target vehicle to identify whether target vehicle drives in the wrong direction and based on mesh
The motion profile growth length of vehicle is marked to identify the driving status of target vehicle, it is as follows to identify rule:
A is about retrograde identification:
Judge whether the motion profile growing method and retrograde direction are identical, if they are the same, corresponding target vehicle occurs inverse
Row;If not identical, there is no driving in the wrong direction.
B: identify about driving status:
Firstly, the position acquisition target detection frame of the target detection frame based on target vehicle in continuous N1 frame image is following
The motion profile growth length of central point;Then, judge whether the motion profile growth length is greater than preset first threshold value L1;
If more than and there is no retrograde, the target vehicle normally travel;
Otherwise, the position acquisition target detection frame of the target detection frame based on target vehicle in continuous N2 frame image is following
The new motion profile growth length of central point, and judge whether the new motion profile growth length is less than default second threshold
L2;
Wherein, N2 is greater than N1, and L2 is less than L1;
If being less than default second threshold L2, judge that current same period same driving direction whether there is other target carriages
?;
If there are other target vehicles, other target vehicles of same driving direction of same period are obtained in continuous N1 frame
Motion profile growth length;
Wherein, if motion profile growth length of other target vehicles in continuous N1 frame is greater than preset first threshold value L1,
The target vehicle is in abnormality, for example scram.If motion profile of other target vehicles in continuous N1 frame is raw
Length is respectively less than default second threshold L2, and the target vehicle is in blocked state;If other targets are in continuous N1 frame
Motion profile growth length is less than or equal to preset first threshold value L1 and is greater than or equal to second threshold L2, at the target vehicle
In state undetermined;
If the target vehicle is in abnormality without other target vehicles;
If the new motion profile growth length is not less than default second threshold L2, the target vehicle is in shape undetermined
State.Wherein state undetermined is the state for needing manually to identify.
Shown in sum up, the present invention provides a kind of automatic means for identifying vehicle and driving in the wrong direction through the above way, while also right
The driving status of vehicle is identified, its application range is extended.
It is emphasized that example of the present invention be it is illustrative, without being restrictive, thus the present invention it is unlimited
Example described in specific embodiment, other all obtained according to the technique and scheme of the present invention by those skilled in the art
Embodiment does not depart from present inventive concept and range, and whether modification or replacement, also belong to protection model of the invention
It encloses.
Claims (8)
- The intelligent detecting method 1. a kind of vehicle based on pursuit path analysis drives in the wrong direction, characterized by the following steps:S1: retrograde detection zone and retrograde bearing mark on camera shooting area image are obtained and obtains camera in real time Video data;Wherein, detection zone of driving in the wrong direction is the quadrilateral area comprising vehicle driving lane, and retrograde approach is labeled as having direction line segment;S2: first-level class device and secondary classifier based on off-line training are inverse when extracting in previous frame image from video data Target vehicle region in row detection zone;Wherein, Classification and Identification is carried out to current frame image using first-level class device and obtains target vehicle candidate region, recycle two Grade classifier carries out Classification and Identification to the target vehicle candidate region and obtains the target vehicle region on current frame image;Wherein, the input data of the first-level class device and the secondary classifier off-line training process is positive sample figure respectively It include vehicle in the positive sample image as, negative sample image, the negative sample image does not include vehicle;S3: correlation filtering tracker is distributed based on the target vehicle region that step S2 is extracted, each nuclear phase closes filter tracker base Other frame image trace target vehicles obtain corresponding target carriage in target vehicle region and video data in current frame image Tracking information;One nuclear phase closes filter tracker and tracks a target vehicle, and current all nuclear phases close filter tracker and corresponding target carriage Constitute current tracking queue;Wherein, other frame images in video data are successively handled according to step S2 and step S3, and known if it exists in treatment process Not Chu target vehicle not in current tracking queue, the target vehicle is fresh target vehicle, and distributes nuclear phase and close filter Wave tracker is for tracking the fresh target vehicle;S4: tracking information and retrograde bearing mark based on target vehicle identify whether target vehicle is sent out in retrograde detection zone It is raw to drive in the wrong direction;Wherein, tracking information includes target detection frame position of the target vehicle in sequential frame image, the target detection frame table Show target vehicle in the tracing area of each frame image;Target detection frame position acquisition based on target vehicle in sequential frame image The motion profile direction of growth of same position on target detection frame, and judge that the motion profile growing method and retrograde direction are No identical, if they are the same, corresponding target vehicle occurs to drive in the wrong direction;If not identical, there is no driving in the wrong direction.
- 2. according to the method described in claim 1, it is characterized by: further including based on target vehicle if vehicle does not occur to drive in the wrong direction Tracking information identify the driving status of target vehicle, process is as follows:Firstly, based on target vehicle on the target detection frame position acquisition target detection frame in continuous N1 frame image same position Motion profile growth length;Then, judge whether the motion profile growth length is greater than preset first threshold value L1;If more than preset first threshold value L1, the target vehicle normally travel;Otherwise, based on target vehicle on the target detection frame position acquisition target detection frame in continuous N2 frame image same position New motion profile growth length, and judge whether the new motion profile growth length is less than default second threshold L2;Wherein, N2 is greater than N1, and L2 is less than L1;If being less than default second threshold L2, judge that current same period same driving direction whether there is other target vehicles;If there are other target vehicles, movement of other target vehicles of same driving direction of same period in continuous N1 frame is obtained Track growth length;Wherein, if motion profile growth length of other target vehicles in continuous N1 frame is greater than default first threshold Value L1, the target vehicle are in abnormality;If motion profile growth length of other target vehicles in continuous N1 frame is equal Less than default second threshold L2, the target vehicle is in blocked state;If motion profile of other targets in continuous N1 frame Growth length is less than or equal to preset first threshold value L1 and is greater than or equal to second threshold L2, and the target vehicle is in shape undetermined State;If the target vehicle is in abnormality without other target vehicles;If the new motion profile growth length is not less than default second threshold L2, the target vehicle is in state undetermined.
- 3. according to the method described in claim 2, it is characterized by: the corresponding preset first threshold value L1 of target vehicle, default the The value range of two threshold value L2 is respectively as follows: L1>2/3h, L2<1/3h;Wherein, h is pixel value shared by the height of target detection frame when tracker captures the target vehicle for the first time.
- 4. according to the method described in claim 2, it is characterized by: in step S4 based on target vehicle in sequential frame image The motion profile direction of growth and length of the following center position of position acquisition target detection frame of target detection frame;Wherein, at interval of the coordinate value of the following center position of n1 frame recording target detection frame.
- 5. according to the method described in claim 1, it is characterized by: further including following steps when handling next frame image:Firstly, tracing area where the target vehicle in tracking queue current in next frame image is carried out mask;Secondly, closing filter tracker using the nuclear phase of target vehicle in current tracking queue obtains target vehicle in next frame image In tracing area;Then, according to the next frame image after step S2 and step S3 processing mask.
- 6. according to the method described in claim 1, it is characterized by: closing filter tracker using nuclear phase obtains corresponding target vehicle Tracking information when, further include supervising whether target vehicle is disappearing within sweep of the eye or reaching the boundary bit of retrograde detection zone It sets;Wherein, if reaching the boundary position of retrograde detection zone in disappearance or target vehicle within sweep of the eye, in currently tracking team The nuclear phase filter tracker is closed in column to delete.
- 7. according to the method described in claim 1, it is characterized by: closing filter tracker using nuclear phase obtains corresponding target vehicle Tracking information when, further include to nuclear phase close filter tracker tracking carry out punishment score;Wherein, filter tracker is closed based on nuclear phase and obtains tracing area of the target vehicle in next frame image, and using in advance Whether the vehicle classification device recognition and tracking region of off-line training includes vehicle;If not including vehicle, filter tracker is closed depending on nuclear phase Tracking error, carries out punishment score to it;Wherein, include vehicle when Probability p is more than or equal to 0.5, when Probability p is less than 0.5, do not include vehicle;Vehicle classification device Obtain be vehicle probability be p when, then score be 1-p;Wherein, if the accumulation score that nuclear phase closes filter tracker reaches default S points, the nuclear phase is closed in current tracking queue Filter tracker is deleted;The input data of the vehicle classification device training process is positive sample image and negative sample image.
- 8. according to the method described in claim 1, it is characterized by: the first-level class device is based on cascade model AdaBoost classifier;The secondary classifier is support vector machines;Wherein, the training process of the first-level class device is trained based on the Haar_like feature of positive sample and negative sample;The training process of the secondary classifier is trained based on the feature vector that positive sample and the HOG feature of negative sample are constituted 's.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910244210.8A CN109948582B (en) | 2019-03-28 | 2019-03-28 | Intelligent vehicle reverse running detection method based on tracking trajectory analysis |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910244210.8A CN109948582B (en) | 2019-03-28 | 2019-03-28 | Intelligent vehicle reverse running detection method based on tracking trajectory analysis |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109948582A true CN109948582A (en) | 2019-06-28 |
CN109948582B CN109948582B (en) | 2021-03-02 |
Family
ID=67012282
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910244210.8A Active CN109948582B (en) | 2019-03-28 | 2019-03-28 | Intelligent vehicle reverse running detection method based on tracking trajectory analysis |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109948582B (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110472496A (en) * | 2019-07-08 | 2019-11-19 | 长安大学 | A kind of traffic video intelligent analysis method based on object detecting and tracking |
CN110473405A (en) * | 2019-09-25 | 2019-11-19 | 拉扎斯网络科技(上海)有限公司 | Driving status detection method, device, readable storage medium storing program for executing and electronic equipment |
CN110555423A (en) * | 2019-09-09 | 2019-12-10 | 南京东控智能交通研究院有限公司 | Multi-dimensional motion camera-based traffic parameter extraction method for aerial video |
CN111046797A (en) * | 2019-12-12 | 2020-04-21 | 天地伟业技术有限公司 | Oil pipeline warning method based on personnel and vehicle behavior analysis |
CN111126171A (en) * | 2019-12-04 | 2020-05-08 | 江西洪都航空工业集团有限责任公司 | Vehicle reverse running detection method and system |
CN111259868A (en) * | 2020-03-10 | 2020-06-09 | 北京以萨技术股份有限公司 | Convolutional neural network-based method, system and medium for detecting vehicles in reverse driving |
CN111354021A (en) * | 2020-02-14 | 2020-06-30 | 广东工业大学 | Target tracking method based on target identification and pixel marking |
CN111611901A (en) * | 2020-05-15 | 2020-09-01 | 北京百度网讯科技有限公司 | Vehicle reverse running detection method, device, equipment and storage medium |
CN111833598A (en) * | 2020-05-14 | 2020-10-27 | 山东科技大学 | Automatic traffic incident monitoring method and system for unmanned aerial vehicle on highway |
CN112507993A (en) * | 2021-02-05 | 2021-03-16 | 上海闪马智能科技有限公司 | Abnormal driving-away detection method and system, electronic equipment and storage medium |
CN112686215A (en) * | 2021-01-26 | 2021-04-20 | 广东工业大学 | Track tracking, monitoring and early warning system and method for carrier loader |
CN112820115A (en) * | 2021-01-05 | 2021-05-18 | 四川铁投信息技术产业投资有限公司 | Method for identifying steering state of running vehicle |
CN113327414A (en) * | 2020-02-28 | 2021-08-31 | 深圳市丰驰顺行信息技术有限公司 | Vehicle reverse running detection method and device, computer equipment and storage medium |
CN113570877A (en) * | 2021-06-22 | 2021-10-29 | 淮阴工学院 | Non-motor vehicle retrograde detection device and detection method |
CN117523521A (en) * | 2024-01-04 | 2024-02-06 | 山东科技大学 | Vehicle detection method based on Haar features and improved HOG features |
Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102768804A (en) * | 2012-07-30 | 2012-11-07 | 江苏物联网研究发展中心 | Video-based traffic information acquisition method |
CN104809887A (en) * | 2015-04-23 | 2015-07-29 | 杭州中威电子股份有限公司 | Detecting method for vehicle converse running on expressway and automatic alarm device |
CN105354529A (en) * | 2015-08-04 | 2016-02-24 | 北京时代云英科技有限公司 | Vehicle converse running detection method and apparatus |
CN105447511A (en) * | 2015-11-15 | 2016-03-30 | 安徽清新互联信息科技有限公司 | SVM object detection method based on Adaboost Haar-Like features |
WO2016080913A1 (en) * | 2014-11-18 | 2016-05-26 | Agency For Science, Technology And Research | Method and device for traffic sign recognition |
US20160342837A1 (en) * | 2015-05-19 | 2016-11-24 | Toyota Motor Engineering & Manufacturing North America, Inc. | Apparatus and method for object tracking |
CN106355884A (en) * | 2016-11-18 | 2017-01-25 | 成都通甲优博科技有限责任公司 | Expressway vehicle guiding system and expressway vehicle guiding method based on vehicle classification |
CN106570439A (en) * | 2015-10-09 | 2017-04-19 | 腾讯科技(深圳)有限公司 | Vehicle detection method and apparatus |
WO2018049514A1 (en) * | 2016-09-13 | 2018-03-22 | Genetec Inc. | Auto-calibration of tracking systems |
US20190012800A1 (en) * | 2017-07-10 | 2019-01-10 | Electronics And Telecommunications Research Institute | Real time multi-object tracking apparatus and method using global motion |
CN109190523A (en) * | 2018-08-17 | 2019-01-11 | 武汉大学 | A kind of automobile detecting following method for early warning of view-based access control model |
-
2019
- 2019-03-28 CN CN201910244210.8A patent/CN109948582B/en active Active
Patent Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102768804A (en) * | 2012-07-30 | 2012-11-07 | 江苏物联网研究发展中心 | Video-based traffic information acquisition method |
WO2016080913A1 (en) * | 2014-11-18 | 2016-05-26 | Agency For Science, Technology And Research | Method and device for traffic sign recognition |
CN104809887A (en) * | 2015-04-23 | 2015-07-29 | 杭州中威电子股份有限公司 | Detecting method for vehicle converse running on expressway and automatic alarm device |
US20160342837A1 (en) * | 2015-05-19 | 2016-11-24 | Toyota Motor Engineering & Manufacturing North America, Inc. | Apparatus and method for object tracking |
CN105354529A (en) * | 2015-08-04 | 2016-02-24 | 北京时代云英科技有限公司 | Vehicle converse running detection method and apparatus |
CN106570439A (en) * | 2015-10-09 | 2017-04-19 | 腾讯科技(深圳)有限公司 | Vehicle detection method and apparatus |
CN105447511A (en) * | 2015-11-15 | 2016-03-30 | 安徽清新互联信息科技有限公司 | SVM object detection method based on Adaboost Haar-Like features |
WO2018049514A1 (en) * | 2016-09-13 | 2018-03-22 | Genetec Inc. | Auto-calibration of tracking systems |
CN106355884A (en) * | 2016-11-18 | 2017-01-25 | 成都通甲优博科技有限责任公司 | Expressway vehicle guiding system and expressway vehicle guiding method based on vehicle classification |
US20190012800A1 (en) * | 2017-07-10 | 2019-01-10 | Electronics And Telecommunications Research Institute | Real time multi-object tracking apparatus and method using global motion |
CN109190523A (en) * | 2018-08-17 | 2019-01-11 | 武汉大学 | A kind of automobile detecting following method for early warning of view-based access control model |
Non-Patent Citations (2)
Title |
---|
HUILING WU等: "Robust online multi-object tracking based on KCF trackers and reassignment", 《2016 IEEE GLOBAL CONFERENCE ON SIGNAL AND INFORMATION PROCESSING》 * |
周行等: "一种基于多特征融合级联分类器的车辆检测算法", 《现代计算机(专业版)》 * |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110472496A (en) * | 2019-07-08 | 2019-11-19 | 长安大学 | A kind of traffic video intelligent analysis method based on object detecting and tracking |
CN110472496B (en) * | 2019-07-08 | 2022-10-11 | 长安大学 | Traffic video intelligent analysis method based on target detection and tracking |
CN110555423A (en) * | 2019-09-09 | 2019-12-10 | 南京东控智能交通研究院有限公司 | Multi-dimensional motion camera-based traffic parameter extraction method for aerial video |
CN110555423B (en) * | 2019-09-09 | 2021-12-21 | 南京东控智能交通研究院有限公司 | Multi-dimensional motion camera-based traffic parameter extraction method for aerial video |
CN110473405B (en) * | 2019-09-25 | 2020-10-23 | 拉扎斯网络科技(上海)有限公司 | Driving state detection method and device, readable storage medium and electronic equipment |
CN110473405A (en) * | 2019-09-25 | 2019-11-19 | 拉扎斯网络科技(上海)有限公司 | Driving status detection method, device, readable storage medium storing program for executing and electronic equipment |
CN111126171A (en) * | 2019-12-04 | 2020-05-08 | 江西洪都航空工业集团有限责任公司 | Vehicle reverse running detection method and system |
CN111046797A (en) * | 2019-12-12 | 2020-04-21 | 天地伟业技术有限公司 | Oil pipeline warning method based on personnel and vehicle behavior analysis |
CN111354021A (en) * | 2020-02-14 | 2020-06-30 | 广东工业大学 | Target tracking method based on target identification and pixel marking |
CN113327414A (en) * | 2020-02-28 | 2021-08-31 | 深圳市丰驰顺行信息技术有限公司 | Vehicle reverse running detection method and device, computer equipment and storage medium |
CN111259868B (en) * | 2020-03-10 | 2023-12-12 | 以萨技术股份有限公司 | Reverse vehicle detection method, system and medium based on convolutional neural network |
CN111259868A (en) * | 2020-03-10 | 2020-06-09 | 北京以萨技术股份有限公司 | Convolutional neural network-based method, system and medium for detecting vehicles in reverse driving |
CN111833598A (en) * | 2020-05-14 | 2020-10-27 | 山东科技大学 | Automatic traffic incident monitoring method and system for unmanned aerial vehicle on highway |
CN111611901B (en) * | 2020-05-15 | 2023-10-03 | 北京百度网讯科技有限公司 | Vehicle reverse running detection method, device, equipment and storage medium |
CN111611901A (en) * | 2020-05-15 | 2020-09-01 | 北京百度网讯科技有限公司 | Vehicle reverse running detection method, device, equipment and storage medium |
CN112820115B (en) * | 2021-01-05 | 2021-11-23 | 四川铁投信息技术产业投资有限公司 | Method for identifying steering state of running vehicle |
CN112820115A (en) * | 2021-01-05 | 2021-05-18 | 四川铁投信息技术产业投资有限公司 | Method for identifying steering state of running vehicle |
CN112686215B (en) * | 2021-01-26 | 2023-07-25 | 广东工业大学 | Track tracking monitoring and early warning system and method for carrier vehicle |
CN112686215A (en) * | 2021-01-26 | 2021-04-20 | 广东工业大学 | Track tracking, monitoring and early warning system and method for carrier loader |
CN112507993A (en) * | 2021-02-05 | 2021-03-16 | 上海闪马智能科技有限公司 | Abnormal driving-away detection method and system, electronic equipment and storage medium |
CN113570877A (en) * | 2021-06-22 | 2021-10-29 | 淮阴工学院 | Non-motor vehicle retrograde detection device and detection method |
CN117523521A (en) * | 2024-01-04 | 2024-02-06 | 山东科技大学 | Vehicle detection method based on Haar features and improved HOG features |
CN117523521B (en) * | 2024-01-04 | 2024-04-02 | 山东科技大学 | Vehicle detection method based on Haar features and improved HOG features |
Also Published As
Publication number | Publication date |
---|---|
CN109948582B (en) | 2021-03-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109948582A (en) | A kind of retrograde intelligent detecting method of vehicle based on pursuit path analysis | |
CN102542289B (en) | Pedestrian volume statistical method based on plurality of Gaussian counting models | |
CN103246896B (en) | A kind of real-time detection and tracking method of robustness vehicle | |
CN105303191A (en) | Method and apparatus for counting pedestrians in foresight monitoring scene | |
CN105512640A (en) | Method for acquiring people flow on the basis of video sequence | |
CN105654508A (en) | Monitoring video moving target tracking method based on self-adaptive background segmentation and system thereof | |
CN110276255A (en) | Based on GMM model and the identification of the high-speed railway foreign matter of BP neural network and classification method | |
CN104134078B (en) | Automatic selection method for classifiers in people flow counting system | |
CN108986472A (en) | One kind turns around vehicle monitoring method and device | |
Kamthe et al. | Suspicious activity recognition in video surveillance system | |
CN103020986A (en) | Method for tracking moving object | |
Makhmutova et al. | Object tracking method for videomonitoring in intelligent transport systems | |
Chang et al. | Video analytics in smart transportation for the AIC'18 challenge | |
CN109829382A (en) | The abnormal object early warning tracing system and method for Behavior-based control feature intelligent analysis | |
CN107103303A (en) | A kind of pedestrian detection method based on GMM backgrounds difference and union feature | |
CN102768726A (en) | Pedestrian detection method for preventing pedestrian collision | |
CN104063692A (en) | Method and system for pedestrian positioning detection | |
CN113963399A (en) | Personnel trajectory retrieval method and device based on multi-algorithm fusion application | |
CN106023650A (en) | Traffic intersection video and computer parallel processing-based real-time pedestrian early-warning method | |
Kumar et al. | A novel approach for speed estimation along with vehicle detection counting | |
Liu et al. | Multi-view vehicle detection and tracking in crossroads | |
CN111104554A (en) | Automatic door control method and system based on contour detection and head and shoulder detection | |
CN115169673A (en) | Intelligent campus epidemic risk monitoring and early warning system and method | |
Bhat et al. | A fully automatic feature-based real-time traffic surveillance system using data association in the probabilistic framework | |
Barnwal | Vehicle Behavior analysis for uneven road surface detection |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |