CN116416565A - Method and system for detecting pedestrian trailing and crossing in specific area - Google Patents
Method and system for detecting pedestrian trailing and crossing in specific area Download PDFInfo
- Publication number
- CN116416565A CN116416565A CN202111630832.8A CN202111630832A CN116416565A CN 116416565 A CN116416565 A CN 116416565A CN 202111630832 A CN202111630832 A CN 202111630832A CN 116416565 A CN116416565 A CN 116416565A
- Authority
- CN
- China
- Prior art keywords
- target
- pedestrian
- specific area
- trailing
- sequence
- 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
- 238000000034 method Methods 0.000 title claims description 22
- 238000001514 detection method Methods 0.000 claims abstract description 122
- 238000012805 post-processing Methods 0.000 claims abstract description 21
- 238000012216 screening Methods 0.000 claims description 30
- 238000012544 monitoring process Methods 0.000 claims description 17
- 238000004458 analytical method Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 5
- 238000012806 monitoring device Methods 0.000 claims description 3
- 238000007689 inspection Methods 0.000 claims description 2
- 230000006399 behavior Effects 0.000 description 21
- 230000008569 process Effects 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 230000033001 locomotion Effects 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 206010000117 Abnormal behaviour Diseases 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Closed-Circuit Television Systems (AREA)
Abstract
The invention provides a detection method for pedestrian trailing and crossing in a specific area, which comprises the steps of obtaining an image sequence; the image sequence carries out target detection and carries out multistage post-processing of a specific area; and carrying out alarm pushing according to the multi-stage post-processing result. The pedestrian trailing and crossing detection method for the specific region provided by the invention can not only utilize target detection as a front-stage detector and introduce the constraint of the position of the specific region, but also effectively help the model to directly pay attention to the key behavior occurrence region, remove complex background interference and improve the detection precision.
Description
Technical Field
The invention relates to the field of pedestrian abnormal behavior detection methods in monitoring scenes, in particular to a detection method and system for pedestrian trailing and crossing in a specific area.
Background
In the monitoring security, the key areas often need to be controlled, and whether suspicious target behaviors exist in the areas or not is detected. In order to locate the true suspicious target behavior from the targets in the area, the targets need to be detected first, and are identified by space-time rule constraint analysis, so as to judge whether the targets accord with the patterns of the suspicious target behavior.
The target detection and identification in the image has wide practical requirements in application occasions such as intelligent video monitoring and the like, and is a popular research direction in the field of computer vision. Mature target detection algorithms can be largely divided into two categories: (1) background-based modeling. The method is mainly used for detecting moving objects in videos: the method comprises the steps of performing scene segmentation on an input static image, segmenting a foreground and a background of the static image by using a Gaussian Mixture Model (GMM) or motion detection method and the like, and extracting a specific moving target from the foreground. Such methods require a continuous sequence of images to achieve modeling and are not suitable for target detection in a single image. (2) statistical learning based. I.e. all images known to belong to a certain class of objects are collected to form a training set, and features are extracted from the training set images based on an artificially designed algorithm (e.g. HOG, harr, etc.). The extracted features are generally information of gray level, texture, gradient histogram, edge and the like of the target. The pedestrian detection classifier is then constructed from a feature library of a large number of training samples. The classifier can be generally a model such as SVM, adaboost and neural network.
In recent years, with the development of deep learning, a target detection algorithm based on statistical learning has made a great breakthrough. In general, the statistical learning-based target detection algorithm can be classified into a conventional artificial feature target detection algorithm and a deep feature machine learning target detection algorithm.
The traditional artificial feature target detection algorithm mainly means that the artificial feature target detection algorithm utilizes the characteristics of artificial design to model target detection. The feature algorithms of artificial design that have been excellent in recent years mainly include: the DPM (Deformable Part Model) algorithm proposed by Pedro F.Felzenszwalb et al in 2010 (Object detection with discriminatively trained part-based models). The Informated Harr method (Informated Haar-like Features Improve Pedestrian Detection) proposed by Shanshan Zhang et al in 2014 is directed to extracting Harr features with more characterization information for training. Although the characteristics of the manual design achieve a certain effect, the detection accuracy is not high because of the insufficient characteristic capacity of the manual characteristics. Because of the stronger feature learning and expression capability of the deep convolutional neural network model, the deep convolutional neural network model is widely applied to the aspect of image target classification detection. The underlying object detection operator is the R-CNN (Region-Convolutional Neural Network) model. In 2014, girshick et al proposed RCNN for detecting general targets, and then proposed fast-RCNN and fast-RCNN, which improved the accuracy and speed of the target detection algorithm based on deep learning.
The suspicious target detection and identification in the monitoring security has wide practical requirements in application occasions such as trailing detection and the like, and is also a popular research direction in the field of computer vision. The object detection based on the deep learning technology is mostly focused on the object detection of a specific image, namely, the positions of various objects are detected from a still picture through appearance characteristics and classified, and the object position and behavior are not analyzed and modeled, so that whether the object is a suspicious object cannot be judged.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a method and a system for detecting pedestrian trailing and crossing in a specific area.
According to one aspect of the present invention, there is provided a pedestrian trailing and crossing detection method for a specific area, including:
acquiring an image sequence;
the image sequence carries out target detection and carries out multistage post-processing of a specific area;
and carrying out alarm pushing according to the multi-stage post-processing result.
Preferably, the target detection includes:
performing frame-by-frame feature extraction on an image sequence obtained by decoding the video monitoring equipment;
and detecting pedestrians frame by frame to obtain a position frame sequence of a pedestrian target.
Preferably, the multi-stage post-treatment comprises:
the position frame sequence of the pedestrian target is used for judging and screening targets in a specific area to obtain the number of the targets;
when the number of the targets is multiple, carrying out track sequence discriminant analysis on the moving tracks of the targets, and analyzing the pedestrian attributes to screen false detection;
and when the target number is a single person, analyzing the attribute of the pedestrian to detect the crossing.
Preferably, the specific area discrimination and screening targets are performed on the position frame sequence of the pedestrian targets, and the target number is obtained, including:
the pedestrian target position frame sequence obtained by target detection and specific region coordinate information aiming at each video monitoring device are utilized to carry out frame-by-frame inspection, and pedestrian targets except the specific region are screened according to the space constraint condition of the specific region;
according to the target quantity, the two situations of multiple persons and single person are respectively sent into different detection flows.
Preferably, when the number of the targets is multiple people, performing a track sequence discriminant analysis on the moving track of the targets, and analyzing the attribute of the pedestrians to screen false detection, including:
extracting position information of an image sequence detected through specific region discrimination;
generating a coordinate sequence corresponding to the target track;
analyzing and checking the coordinate sequence of the target track by utilizing speed and direction constraint, and judging whether the target track belongs to trailing;
child trailing events that are determined to be trailing for both the specific region discrimination and the trajectory sequence discrimination,
and acquiring an image of the pedestrian by cutting the original image, judging the attribute of the pedestrian, identifying the attribute of the target age and screening out the tail of the child.
Preferably, when the target number is a single person, analyzing the pedestrian attribute to detect a walk-over includes:
cutting the image sequence which is detected through the specific area discrimination to obtain a target image;
identifying the age and attitude attribute of the target image;
and if the judging gesture is overturned, the corresponding alarm is given.
According to a second aspect of the present invention, there is provided a specific area pedestrian trailing and crossing detection system comprising
The target detection module is used for positioning a pedestrian target;
the multi-stage post-processing module is used for processing the specific area of the pedestrian target; the multi-stage post-processing module comprises:
the specific area judging module receives the image sequence pushed by the target detection module and deletes irrelevant targets in the image sequence; extracting target positions in the image sequence, and judging the number of targets in the specific area;
the track sequence judging module judges whether the target number is trailing when the target number is multiple persons;
and the pedestrian attribute judging module judges the attribute of the target in the image sequence and performs false detection screening and crossing analysis.
Preferably, the target detection module and the specific area discriminating module include:
extracting an image sequence from the monitoring video, sequentially sending each frame of image to the target detection module, and detecting targets in the images to obtain a target frame sequence;
collecting and traversing all the target frames, and comparing the central point of the target frame with the coordinates of a specific area;
if two or more pedestrian center points fall into the same specific area, the following false detection screening is needed;
if only one pedestrian center point falls into a specific area, the detection judgment of the crossing is needed.
Preferably, the track sequence discriminating module includes:
extracting target position information of each frame of image of an image sequence aiming at the image sequence which is judged to be trailing by the input specific area judging module, and obtaining a target frame center point track sequence;
and judging trailing when the track sequence meets the monotonicity rule, namely that two or more pedestrians in a specific area move in the same direction at a certain speed at the same time.
Preferably, the pedestrian attribute discriminating module includes:
extracting target position information of each frame image of an input image sequence, and cutting an original image according to the target frame coordinates to obtain a pedestrian target image;
and (3) identifying the age and gesture attribute of the pedestrian target image, wherein the attribute comprises trailing false detection screening and crossing detection.
Compared with the prior art, the invention has the following beneficial effects:
the pedestrian trailing and crossing detection method for the specific region provided by the invention can not only utilize target detection as a front-stage detector and introduce the constraint of the position of the specific region, but also effectively help the model to directly pay attention to the key behavior occurrence region, remove complex background interference and improve the detection precision.
In addition, the problem of insufficient information of single-frame image target detection can be solved by mining and utilizing the target coordinate information of adjacent frames and identifying the target attribute.
Drawings
Other features, objects and advantages of the present invention will become more apparent upon reading of the detailed description of non-limiting embodiments, given with reference to the accompanying drawings in which:
FIG. 1 is a flow chart of a method for detecting pedestrian trailing and crossing in a specific area according to an embodiment of the invention;
FIG. 2 is a flow chart of a specific area determining module according to an embodiment of the invention;
FIG. 3 is a flowchart of a track sequence determination module according to an embodiment of the invention;
fig. 4 is a flowchart of a pedestrian attribute determination module according to an embodiment of the invention.
Detailed Description
The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the present invention, but are not intended to limit the invention in any way. It should be noted that variations and modifications could be made by those skilled in the art without departing from the inventive concept. These are all within the scope of the present invention.
Aiming at the defects that the conventional target detection is used for analyzing a still image in isolation and is difficult to judge the action behavior of the target, after multi-frame image information is fused, the behavior of the target is analyzed by utilizing the characteristics of time sequence and space and constraint conditions, so that the interference of the normal target can be filtered, and the real suspicious target is found. In practical application, the method has the advantages that the target detection is used as a front-stage detector, and the constraint of the position of a specific region is introduced, so that the model can be effectively assisted to directly pay attention to the key behavior occurrence region, complex background interference is removed, and the detection precision is improved. In addition, the defects of single-frame image target detection can be effectively overcome by mining and utilizing the target coordinate information of the adjacent frames and identifying the target attribute.
Based on the above-mentioned conception, an embodiment of the present invention is provided, as shown in fig. 1, which is a flowchart of a method for detecting the trailing and crossing of pedestrians in a specific area, including:
s1, acquiring an image sequence;
s2, performing target detection on the image sequence and performing multistage post-processing on a specific area;
and S3, carrying out alarm pushing according to the multi-stage post-processing result.
Based on the above embodiments, a preferred embodiment is provided. Comprising
S100, constructing a model structure of combining a target detection network with multi-stage post-processing.
In this embodiment, a detection network is constructed based on target detection and multi-stage post-processing. The target detection is a target detection model, an image is input, and the target position is detected. The multi-stage post-processing includes specific region discrimination processing, trajectory sequence discrimination processing, and pedestrian attribute discrimination processing.
S200, positioning the target by using the target detection model.
In this embodiment, the current frame image is taken out from the image buffer sequence obtained by decoding the input device, and sequentially sent to the target detection network to perform frame-by-frame feature extraction, and the target frame is screened based on pedestrian category information on the basis of detecting the target, so as to finally obtain the position frame sequence of the pedestrian target for the subsequent processing process.
Specifically, the current frame image is sequentially sent into a YOLOv5 target detection model to extract frame-by-frame characteristics, and multi-scale deep features are fused to finally output frame coordinates, categories and confidence of the target. And (3) utilizing target information predicted by the target detection model, utilizing a confidence threshold and a non-maximum value to inhibit and screen redundant candidate frames on the basis of detecting the target, and screening target frames based on pedestrian category information, so as to obtain a pedestrian target position frame sequence finally used for judging the trailing behavior.
S300, screening targets to be focused by utilizing a post-processing specific area judging module.
In the step, the pedestrian target area obtained by the target detection module and the specific monitoring area coordinate information specific to each input device are utilized to carry out frame-by-frame detection, the focus is concentrated on the core area where the trailing and the crossing occur, the judgment is carried out according to the space constraint condition of the specific area, if a plurality of people are located in the specific area at the same time, the current image sequence is judged to be the suspicious trailing, and if a single person is present, the crossing detection is only carried out.
The specific area is distinguished as a filtering mechanism of the first step, and a plurality of people are tentatively distinguished as trailing events and submitted to track sequence distinguishing for further screening in the specific area. The specific remaining space constraint condition, namely the target position frame, is required to be within a specific monitoring area, specifically, the pedestrian target is required to appear in a designated partial area in the video picture to be reserved, otherwise, the target position frame is determined to be irrelevant to screen out
Specifically, a specific area discriminating module is utilized to screen objects to be focused on: firstly, selecting a specific area position according to corresponding frames of different monitoring scenes, wherein the position frame is required to be attached to an actual specific area. Referring to fig. 2, an image sequence is extracted from a surveillance video at a fixed frame rate, and each frame of image is sequentially sent to a target detector to detect a target existing in the image, thereby obtaining a pedestrian target frame sequence.
And putting all the pedestrian target frames into a set, traversing the pedestrian target frames, firstly averaging according to the coordinate information of the pedestrian target frames to obtain the center point position of the target frame, then comparing the center point position of the target frame with a specific area, and if the center point of the pedestrian target frame falls into the specific area, adding the pedestrian target into a cache queue corresponding to the specific area, otherwise, discarding.
After all the pedestrian targets are processed, all the specific pedestrian target information is taken out from the cache queue. If two or more pedestrian targets appear in the same specific area, the trailing false detection screening is carried out, and if only one person appears, the crossover detection is carried out.
S400, analyzing the target movement track by utilizing a track sequence discrimination module for the multiple-person situation.
In this embodiment, referring to fig. 3, on the basis of the image sequence detected by the specific area discriminating module, the track sequence discriminating module first extracts the position information of the center point of the pedestrian target from the image sequence, and generates a coordinate sequence corresponding to the target track of the suspicious pedestrian marked by the pedestrian, if the coordinate track indicates that multiple people in the specific area move in the specific exit direction, i.e. in the specific direction, at a higher speed, the track sequence discriminating module can determine that the pedestrian belongs to the trailing, otherwise, the pedestrian is misreported and screened out.
S500, screening false detection of multiple people by utilizing a pedestrian attribute judging module.
In this embodiment, referring to fig. 4, the pedestrian attribute discriminating module is used to screen out the trailing false detection for the multiple person situation: referring to fig. 4, for the detected trailing image sequence, the pedestrian attribute discriminating module extracts the position information of the pedestrian target in the specific area of the image sequence, cuts the image of the pedestrian target based on the position information, sends the cut image of the pedestrian target into the attribute classifier to identify the attribute, wherein the image contains age and gesture information, and for the trailing detection branch logic, if the age is too small, the trailing of the child is determined to be screened out.
S600, detecting and crossing by using a pedestrian attribute judging module for single person situations.
Detecting the crossing by using the pedestrian attribute distinguishing module aiming at the single person situation: the method comprises the steps that a cached image sequence provided by a specific area judging module is obtained, a pedestrian attribute judging module firstly extracts position information of pedestrian targets in a specific area of the image sequence, cuts images of the pedestrian targets in the specific area based on position coordinates, sends the cut pedestrian target images into an attribute classifier to identify attributes, wherein the images comprise age and attitude information, and if the logic of a crossing detection branch line appears, the logic is classified as a crossing attitude, and then the logic is judged to be a crossing.
The object detection of the above embodiment is mainly required to solve the following problems:
the angles are different. Due to constraints of actual scene topography, environment and the like, different monitoring devices necessarily have different viewing angles, such as difference in height and difference in depression angle, and different angular views mean that input images have higher differences, and key areas for actual monitoring may be located at different positions in an image sequence.
Subsequent analysis is lacking. The object detection utilizes the appearance characteristics of the object to judge the position and the category of the object, the specific behavior cannot be judged by the coordinate information of the object frame, and additional space-time conditions and characteristic constraints are required to be introduced to define the suspicious object behavior, so that whether the suspicious object belongs to the suspicious object or the normal object is judged.
Based on the difficulty in detecting the target in reality, the method for detecting the pedestrian trailing and crossing in the specific area in the embodiment firstly sends the original image into the target detection module to detect the pedestrian target, and then uses the coordinate position of the specific area to carry out space judgment on the detected target, and the method is divided into two cases of multiple people and single person. If a plurality of pedestrian targets are simultaneously present in a specific area, judging that the images are suspicious trailing sample image sequences; extracting pedestrian position information in the suspicious sample, storing the pedestrian position information in a track sequence, and sending the track sequence into a track judging module, and judging whether the pedestrian position information is trailing according to track data; if the images are determined to be suspicious, extracting pedestrian attribute information in the suspicious samples, screening out child trailing events according to the ages of pedestrians, and pushing the suspicious trailing event image sequences to a background for alarming. If only a single pedestrian target appears in a specific area, only attribute information of the pedestrian is extracted, and whether the pedestrian belongs to the crossing or not is judged by utilizing the attitude attribute. According to the pedestrian trailing and crossing detection method for the specific area, which is provided by the embodiment of the invention, from the problems, the detection problem of the trailing target under the difference of different visual angles can be better solved by a multi-stage post-processing space-time multiple constraint mode, and the behavior of the target can be effectively analyzed, so that whether the behavior is suspicious or not can be judged.
The method provided by the embodiment of the invention comprises the steps of constructing a target detector, screening targets based on specific region definition, false detection screening strategies and pedestrian attribute identification; the whole detection process comprises five links, and is described by taking detection of the tail and crossing behaviors of the railway inbound gate as an example. In the present embodiment, the specific region refers to a partial region in the video picture. The specific area is the area including the gate, which is relative to the whole monitor screen.
S10, monitoring pedestrians at the entrance of the railway by using a monitoring camera installed at the entrance of the railway, maintaining a cache queue of input images in order to ensure that the target detection only detects images at the current time, and only acquiring images shot by the camera from the cache queue by the target detector. And processing the input image by using a YOLOv5 model, wherein the image preprocessing and the reasoning process of extracting image features by a convolution network are included, and the specific position and the target category of the target frame are regressed based on the features. And screening pedestrian targets with high confidence according to the confidence and the category of the targets.
S20, detecting the trailing by using the specific area judging module. Firstly, specific area positions are selected according to different camera corresponding frames, the position frames are attached to the inner side boundaries of the actual specific areas, and the specific entrance marking positions can be properly deepened. The image sequence is extracted from the monitoring video at a certain frame rate, for example 8 frames or 12 frames per second, and each frame of image is sequentially sent to the object detector to detect the object existing in the image, so that the pedestrian object frame sequence is obtained. And putting all the pedestrian target frames detected by the YOLOv5 target detection model into a set, traversing the pedestrian target frames, firstly averaging according to the coordinate information of the pedestrian target frames to obtain the center point position of the target frames, then comparing the center point position of the target frames with a specific area, and if the center point of the pedestrian target frames falls into the specific area, adding the pedestrian target into a buffer queue corresponding to the specific area, otherwise discarding the pedestrian target. After all the pedestrian targets are processed, all the specific pedestrian target information is taken out from the cache queue. If two or more pedestrian targets appear in the same specific area, the suspicious trailing event is judged to exist at the specific area for subsequent false detection screening, and if only a single pedestrian target appears in the same specific area, the judgment is carried out.
S30, analyzing the target movement track by utilizing a track sequence discrimination module for the multiple-person situation. For the trailing image sequence detected by the specific area judging module, firstly extracting the position information of the center point of the pedestrian target with suspicious trailing behaviors from the image sequence acquired by the corresponding camera, generating a coordinate sequence of the target track of the suspicious pedestrian marked correspondingly, wherein each sequence needs to correspond to a specific number under the current monitoring input, if the coordinate track shows that two persons in the specific area move in the direction of a specific exit, namely in the specific direction, at a speed exceeding a certain threshold value, judging the position of the pedestrian target with suspicious trailing behaviors, otherwise marking the position as false alarm and screening out the false alarm. The actual specific direction may be changed due to different camera angles and algorithm speeds, the speed direction and threshold parameters need to be adjusted according to actual conditions, the speed threshold takes pixel points as a unit, and the reference value is 10px under the proper camera angles and the processing speed of 8 frames of images per second.
S40, screening false child detection by using a pedestrian attribute judging module for the multiple-person situation. And firstly extracting position information of the detected pedestrian target with suspicious trailing behaviors, cutting an original image on the basis of the position information to obtain a pedestrian image, taking the pedestrian target image as input of an attribute extraction classifier, obtaining age attributes of the target through identification and classification, and screening out the suspicious trailing image sequence if the age attributes are judged to be children.
S50, detecting and crossing the single person by using the pedestrian attribute judging module. The method comprises the steps of firstly extracting pedestrian target position information from a cache image sequence provided by a specific region judging module, cutting an original image on the basis of the position information to obtain a pedestrian image, taking the pedestrian target image as input of an attribute extraction classifier, obtaining the attitude attribute of a target through identification and classification, and judging to be overturned and giving a corresponding alarm if the attitude is abnormal, namely classifying to be overturned.
In the embodiment of the invention, the following and crossing behavior refers to the behavior of illegally following other people or crossing to pass in a specific monitored area, for example, at a specific place of a railway station entrance, the following other people or crossing a gate to realize ticket escaping or avoid the illegal purpose of identity verification.
According to the method, a target detection model is used for detecting a pedestrian target area, predefined specific area coordinates are used for screening pedestrian targets, a multi-person situation image sequence is sent to a track discrimination and attribute identification module for false detection screening, and a single person situation image sequence is sent to the attribute identification module for crossing detection, so that interference of a camera angle and a specific area position is effectively avoided, suspicious behaviors of the targets are analyzed, and detection of trailing and crossing behaviors in a video is achieved.
Based on the same conception of the above embodiments, in other embodiments of the present invention, a specific area pedestrian trailing and crossing detection system is also provided, including a target detection module and a multi-stage post-processing module, where the target detection module locates a pedestrian target; the multistage post-processing module processes the pedestrian target in a specific area; the multi-stage post-processing module comprises: the specific area judging module receives the image sequence pushed by the target detecting module and deletes irrelevant targets in the image sequence to screen out; extracting target positions in the image sequence, and judging the number of targets in the specific area; the track sequence judging module judges whether the target number is trailing when the target number is multiple persons; and the pedestrian attribute judging module judges the attribute of the target in the image sequence and performs false detection screening and crossing analysis.
Further, the target detection module and the specific area discrimination module include:
extracting an image sequence from the monitoring video, sequentially sending each frame of image to the target detection module, and detecting targets in the images to obtain a target frame sequence;
collecting and traversing all the target frames, and comparing the central point of the target frame with the coordinates of a specific area;
if two or more pedestrian center points fall into the same specific area, the following false detection screening is needed;
if only one pedestrian center point falls into a specific area, the detection judgment of the crossing is needed.
Further, the track sequence discriminating module includes:
extracting target position information of each frame of image of an image sequence aiming at the image sequence which is judged to be trailing by the input specific area judging module, and obtaining a target frame center point track sequence;
and judging trailing when the track sequence meets the monotonicity rule, namely that two or more pedestrians in a specific area move in the same direction at a certain speed at the same time.
Further, the pedestrian attribute discriminating module includes:
extracting target position information of each frame image of an input image sequence, and cutting an original image according to the target frame coordinates to obtain a pedestrian target image;
and (3) identifying the age and gesture attribute of the pedestrian target image, wherein the attribute comprises trailing false detection screening and crossing detection.
The foregoing describes specific embodiments of the present invention. It is to be understood that the invention is not limited to the particular embodiments described above, and that various changes and modifications may be made by one skilled in the art within the scope of the claims without affecting the spirit of the invention. The above-described preferred features may be used in any combination without collision.
Claims (10)
1. The pedestrian trailing and crossing detection method for the specific area is characterized by comprising the following steps of:
acquiring an image sequence;
the image sequence carries out target detection and carries out multistage post-processing of a specific area;
and carrying out alarm pushing according to the multi-stage post-processing result.
2. The specific area pedestrian trailing and crossing detection method according to claim 1, wherein the target detection includes:
performing frame-by-frame feature extraction on an image sequence obtained by decoding the video monitoring equipment;
and detecting pedestrians frame by frame to obtain a position frame sequence of a pedestrian target.
3. The specific area pedestrian trailing and crossing detection method according to claim 2, wherein the multi-stage post-processing includes:
the position frame sequence of the pedestrian target is used for judging and screening targets in a specific area to obtain the number of the targets;
when the number of the targets is multiple, judging the moving track of the targets by a track sequence, and analyzing the attribute of the pedestrians to screen out false detection;
and when the target number is a single person, analyzing the attribute of the pedestrian to detect the crossing.
4. The method for detecting pedestrian trailing and crossing in a specific area according to claim 3, wherein the sequence of position frames of pedestrian targets performs specific area discrimination screening targets to obtain the number of targets, comprising:
performing frame-by-frame inspection by utilizing a position frame sequence of the pedestrian targets obtained by target detection and specific region coordinate information of each video monitoring device, and screening the pedestrian targets except the specific region according to space constraint conditions of the specific region to obtain the number of the pedestrian targets in the specific region;
according to the target quantity, the two situations of multiple persons and single person are respectively sent into different detection flows.
5. The method for detecting pedestrian trailing and crossing in a specific area according to claim 4, wherein when the number of the targets is a plurality of persons, performing a trajectory sequence discriminant analysis of the moving trajectories of the targets and analyzing pedestrian attributes to screen out false detection, comprising:
extracting position information of the image sequence detected through the specific region discrimination, and generating a coordinate sequence corresponding to the target track;
analyzing and checking the coordinate sequence of the target track by utilizing speed and direction constraint, and judging whether the target track belongs to trailing;
aiming at child trailing events which are judged to be trailing by specific region judgment and track sequence judgment, the original image is cut, pedestrian attribute judgment is carried out on the images of pedestrians, target age attributes are identified, and child trailing is screened out.
6. The specific area pedestrian trailing and crossing detection method according to claim 4, wherein when the target number is a single person, analyzing pedestrian attributes to detect crossing comprises:
cutting the image sequence which is detected through the specific area discrimination to obtain a target image;
identifying the age and attitude attribute of the target image;
and if the gesture is judged to be overturned, alarming.
7. A pedestrian trailing and crossing detection system in a specific area is characterized by comprising
The target detection module is used for positioning a pedestrian target;
the multi-stage post-processing module is used for processing the specific area of the pedestrian target; the multi-stage post-processing module comprises:
the specific area judging module receives the image sequence pushed by the target detection module and deletes irrelevant targets in the image sequence; extracting target positions in the image sequence, and judging the number of targets in the specific area;
the track sequence judging module judges whether the target number is trailing when the target number is multiple persons;
and the pedestrian attribute judging module judges the attribute of the target in the image sequence and performs false detection screening and crossing analysis.
8. The zone-specific pedestrian trailing and crossing detection system of claim 7, wherein the target detection module and zone-specific discrimination module comprise:
extracting an image sequence from the monitoring video, sequentially sending each frame of image to the target detection module, and detecting targets in the images to obtain a target frame sequence;
traversing after collecting all target frames, and comparing the central point of the target frame with the coordinates of a specific area;
if two or more pedestrian center points fall into the same specific area, the following false detection screening is needed;
if only one pedestrian center point falls into a specific area, the detection judgment of the crossing is needed.
9. The specific area pedestrian trailing and crossing detection system of claim 7, wherein the track sequence determination module comprises:
extracting target position information of each frame of image of an image sequence aiming at the image sequence which is judged to be trailing by the input specific area judging module, and obtaining a target frame center point track sequence;
and judging trailing when the track sequence meets the monotonicity rule, namely that two or more pedestrians in a specific area move in the same direction at a certain speed at the same time.
10. The specific area pedestrian trailing and crossing detection system of claim 7, wherein the pedestrian attribute determination module comprises:
extracting target position information of each frame image of an input image sequence, and cutting an original image according to the target frame coordinates to obtain a pedestrian target image;
and (3) identifying the age and gesture attribute of the pedestrian target image, wherein the attribute comprises trailing false detection screening and crossing detection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111630832.8A CN116416565A (en) | 2021-12-29 | 2021-12-29 | Method and system for detecting pedestrian trailing and crossing in specific area |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111630832.8A CN116416565A (en) | 2021-12-29 | 2021-12-29 | Method and system for detecting pedestrian trailing and crossing in specific area |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116416565A true CN116416565A (en) | 2023-07-11 |
Family
ID=87056388
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111630832.8A Pending CN116416565A (en) | 2021-12-29 | 2021-12-29 | Method and system for detecting pedestrian trailing and crossing in specific area |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116416565A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118038340A (en) * | 2024-04-15 | 2024-05-14 | 盛视科技股份有限公司 | Anti-trailing detection system based on video image |
-
2021
- 2021-12-29 CN CN202111630832.8A patent/CN116416565A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118038340A (en) * | 2024-04-15 | 2024-05-14 | 盛视科技股份有限公司 | Anti-trailing detection system based on video image |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11188783B2 (en) | Reverse neural network for object re-identification | |
| CN106203274B (en) | Real-time pedestrian detection system and method in video monitoring | |
| CN113011367B (en) | Abnormal behavior analysis method based on target track | |
| Yuan et al. | Statistical hypothesis detector for abnormal event detection in crowded scenes | |
| Stalder et al. | Cascaded confidence filtering for improved tracking-by-detection | |
| US9569531B2 (en) | System and method for multi-agent event detection and recognition | |
| TW201737134A (en) | System and method for training object classifier by machine learning | |
| US20060067562A1 (en) | Detection of moving objects in a video | |
| KR101653278B1 (en) | Face tracking system using colar-based face detection method | |
| San Miguel et al. | Robust unattended and stolen object detection by fusing simple algorithms | |
| CN105303191A (en) | Method and apparatus for counting pedestrians in foresight monitoring scene | |
| CN105893946A (en) | Front face image detection method | |
| Piciarelli et al. | Surveillance-oriented event detection in video streams | |
| CN110490043A (en) | A kind of forest rocket detection method based on region division and feature extraction | |
| KR101472674B1 (en) | Method and apparatus for video surveillance based on detecting abnormal behavior using extraction of trajectories from crowd in images | |
| CN106570490A (en) | Pedestrian real-time tracking method based on fast clustering | |
| WO2013075295A1 (en) | Clothing identification method and system for low-resolution video | |
| CN113963399A (en) | Personnel trajectory retrieval method and device based on multi-algorithm fusion application | |
| CN110826390A (en) | Video data processing method based on face vector characteristics | |
| CN111222449A (en) | Driver behavior detection method based on fixed camera image | |
| CN116416565A (en) | Method and system for detecting pedestrian trailing and crossing in specific area | |
| CN103971100A (en) | Video-based camouflage and peeping behavior detection method for automated teller machine | |
| CN110502995B (en) | Driver yawning detection method based on fine facial action recognition | |
| Wei et al. | Target-tracking based early fire smoke detection in video | |
| CN111104554A (en) | Automatic door control method and system based on contour detection and head and shoulder 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 |