WO2023071397A1 - Detection method and system for dangerous driving behavior - Google Patents
Detection method and system for dangerous driving behavior Download PDFInfo
- Publication number
- WO2023071397A1 WO2023071397A1 PCT/CN2022/111853 CN2022111853W WO2023071397A1 WO 2023071397 A1 WO2023071397 A1 WO 2023071397A1 CN 2022111853 W CN2022111853 W CN 2022111853W WO 2023071397 A1 WO2023071397 A1 WO 2023071397A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- target
- driving behavior
- dangerous driving
- detection
- algorithm based
- Prior art date
Links
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
- G06N3/04—Architecture, e.g. interconnection topology
-
- 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
Definitions
- the invention relates to the technical field of feature recognition, in particular to a detection method and system for dangerous driving behavior.
- the driver’s driving habits can be regulated through means similar to violation penalties, thereby reducing the probability of drivers’ bad driving behaviors, thereby reducing the potential for traffic accidents .
- the number of urban traffic accidents can be reduced. Therefore, it has very important practical value to accurately detect and recognize the driving behavior of road drivers.
- the existing dangerous driving behavior recognition technology can only identify some simple violations for single vehicles, and cannot identify dangerous behaviors between vehicles and between vehicles.
- the violations of bicycles are the potential cause of traffic accidents, the most likely cause of accidents is the dangerous driving behavior caused by the direct interaction of multiple objects in road traffic. For this reason, it is particularly important to identify dangerous driving behaviors other than violations, so that some dangerous driving behaviors that do not violate the rules but may directly cause accidents can be identified and dealt with, which can greatly increase the probability of road traffic accidents. Greatly reduce the loss of human life and property.
- the technical problem to be solved by the present invention is to provide an intelligent detection and processing system for dangerous driving behavior, which can realize the identification of dangerous driving behavior of road drivers, and then through intervention means, and finally reduce the probability of urban road traffic accidents. detection method and system.
- a detection method for dangerous driving behavior comprising:
- the monitoring image sequence is based on the deep learning-based moving target detection algorithm, and the target fast tracking algorithm based on target matching and allocation to obtain the parameter matrix of the moving target;
- the parameter matrix includes target ID, acceleration vector;
- the moving target detection algorithm based on deep learning and the target fast tracking algorithm based on target matching and distribution adopt a tracking by detection framework, wherein detection is responsible for the detection and identification of moving targets, and tracking is responsible for tracking the inter-frame movement of moving targets;
- the target fast tracking algorithm based on target matching and allocation includes Kalman prediction model based on linear motion model, target matching algorithm based on overlap evaluation, and target allocation algorithm based on Hungarian algorithm.
- the Kalman prediction model based on the linear motion model includes:
- the preset moving target moves linearly between two adjacent frames of images, and then predicts the position of the moving target in the next frame of images;
- the state vector includes the position parameter, velocity parameter and the size parameter of the detection frame and the size change parameter of the moving target in the monitoring image sequence, and Gaussian white noise is used as the system noise and measurement noise, and the Kalman prediction equations are set up to output the result .
- the target matching algorithm based on overlap evaluation includes:
- the overlapping degree calculation formula of the overlapping degree evaluation is:
- c is candidatebound, that is, the candidate frame
- g is groundtruthbound, that is, the original mark frame
- area represents the area
- IOU intersectionoverunion, that is, the degree of overlap.
- the target allocation algorithm based on the Hungarian algorithm comprises:
- the moving target allocation problem in two adjacent frames of images is summarized as a bipartite graph problem, and a recursive method is used to iteratively find the augmenting path.
- the moving target detection algorithm based on deep learning classifies and locates targets in the surveillance image sequence.
- a detection system for dangerous driving behavior comprising:
- a number of traffic monitoring devices generating video streams
- the traffic monitoring local area network A has a real-time database server, an on-site workstation and a monitoring workstation, the real-time database server stores the video stream, and the on-site workstation uses the detection method for dangerous driving behavior described in any one of claims 1-7 to detect the video stream ;
- the monitoring workstation retrieves the video stream and displays it in real time;
- the traffic monitoring local area network B mounts the real-time database server, on-site workstation and monitoring workstation of the traffic monitoring local area network A to the traffic monitoring local area network B at the same time, and realizes the distribution of external information through the Web server;
- the receiving terminal receives the external information distributed from the traffic monitoring LAN B.
- the on-site workstation first decodes the video stream, extracts the moving object, and classifies the moving object before using the dangerous driving behavior detection method described in any one of claims 1-7 to detect the video stream .
- the external information includes start and end time, participants, and video;
- the traffic monitoring local area network B also includes a real-time/historical database, and the real-time/historical database stores the information uploaded by the real-time database server of the traffic monitoring local area network A;
- the Web server receives the information output by the field workstation through the traffic monitoring local area network B, and expands the information, and retrieves the place of occurrence of the information and the corresponding video clips through the real-time/historical database to form complete dangerous driving behavior information;
- the terminal user is screened by the receiving terminal based on the occurrence location information, and the external information including complete accident information is sent to the receiving terminal corresponding to the terminal user.
- the beneficial effect of the present invention is that it can intelligently judge whether dangerous driving behavior occurs in road traffic by automatically identifying and processing the motion relationship between objects; it solves the problem of identifying dangerous driving behavior that is likely to cause accidents.
- Traffic control departments or insurance companies can obtain this information at the first time, so that they can carry out education or various restraint behaviors of car owners based on this information, so as to correct the driving habits of drivers and reduce the probability of dangerous driving behaviors as much as possible, thus greatly It is possible to reduce the occurrence of road traffic accidents and achieve the purpose of safe travel.
- Fig. 1 is a schematic flow chart of a detection method for dangerous driving behavior according to a specific embodiment of the present invention
- FIG. 2 is a schematic flowchart of a detection method for dangerous driving behavior based on deep learning and a fast-tracking algorithm based on target matching and distribution in a specific embodiment of the present invention
- FIG. 3 is a schematic diagram of a dangerous driving judgment algorithm of a dangerous driving behavior detection method according to a specific embodiment of the present invention.
- FIG. 4 is a schematic diagram of a detection system for dangerous driving behavior according to a specific embodiment of the present invention.
- FIG. 5 is a schematic diagram of a detection process of a dangerous driving behavior detection system according to a specific embodiment of the present invention.
- FIG. 6 is a schematic diagram of an information processing flow of a dangerous driving behavior detection method according to a specific embodiment of the present invention.
- a detection method for dangerous driving behavior including:
- the parameter matrix of the target is obtained by applying the moving target detection algorithm based on deep learning and the target fast tracking algorithm based on target matching and allocation to the monitoring image sequence, and its parameters include target ID, acceleration vector and other information. Thereafter, by comparing the target ID acceleration vector in the target parameter matrix with the target threshold, it is judged whether dangerous driving behavior occurs.
- the detection method of the dangerous driving behavior uses a tracking by detection framework, wherein detection is responsible for the detection and recognition of moving objects, and tracking is responsible for tracking the inter-frame movement of moving objects, and its specific implementation is as shown in Figure 2.
- the network classifies and localizes objects in a sequence of surveillance images. Its idea is to measure the input information through a multi-layer neural network, and continuously calculate and learn through a large amount of data, so that the machine can learn how to perform tasks.
- the network needs to complete the classification and positioning of 7 types of moving objects such as pedestrians, non-motor vehicles, and motor vehicles. According to the needs, the structural parameters of the network were modified accordingly, the image was divided into 7*7 grids, the grid prediction target frame was set to 5, and the output types were set to 7 categories (pedestrians, people riding two-wheelers, There are seven types of tricycle riders, cars, SUVs, buses, and trucks), so that a 7*7*5*(7+5) dimensional moving target detection result can be obtained through the network.
- Kalman prediction is an algorithm that uses the linear system state equation to optimally estimate the system state through the input state vector and output observation data of the system.
- the state vector involved in the present invention is made up of the position parameter (x, y) of moving object, velocity parameter (dx, dy), and the size parameter (w, h) of detection frame, size variation parameter (dw, dh) of Gaussian
- the white noise is the system noise and the measurement noise, and the Kalman prediction equations are set up.
- the target matching algorithm can correlate the information of the same moving target on two adjacent frames of images, so as to describe the position sequence of the moving target.
- the core of the target matching algorithm is to use the overlapping degree to evaluate the target detection results and the output results of the Kalman prediction model.
- the degree of overlap is a concept often used in target detection. It is the overlap rate between the generated candidate frame and the original marked frame. The calculation is shown in the following formula.
- c is the candidate bound, that is, the candidate frame
- g is the ground truth bound, that is, the original marked frame
- area represents the area
- IOU is the intersection over union, that is, the degree of overlap.
- the present invention summarizes the problem of assigning moving objects in two adjacent frames (k-1 frame, k frame) into a bipartite graph problem, that is, finds pairs for the most moving objects in k-1 frame images as much as possible.
- the present invention establishes a priori rule-based dangerous driving behavior judgment model according to the ID and acceleration vector of the moving object, as shown in FIG. 3 .
- a detection system for dangerous driving behavior including:
- the real-time database server is responsible for storing the video stream in LAN A; the monitoring workstation can call the traffic monitoring equipment in LAN A and display the video in real time; the on-site workstation can process and detect the traffic monitoring video stream in LAN A in real time And extract dangerous driving behavior.
- the real-time database server, on-site workstation and monitoring workstation are mounted to the traffic monitoring LAN B at the same time, and the distribution of external information can be realized through the Web server.
- the receiving terminal of the system receives the message sent by the Web server through the Internet.
- the dangerous driving behavior technology mainly involves processing the video output from the traffic monitoring equipment mounted in the traffic monitoring LAN A, and automatically detecting and extracting dangerous driving behavior data from it, as shown in Figure 5.
- the traffic monitoring video stream is decoded.
- the deep neural network YOLOv3 uses the deep neural network YOLOv3 to extract moving objects.
- moving objects There are seven types of moving objects: pedestrians, people riding two-wheelers, people riding tricycles, cars, SUVs, buses, and trucks.
- the tracking of the moving target is carried out, and the parameter matrix of the same moving target in a period of time is output, including ID, position and velocity vector.
- the parameter matrix is input into the dangerous driving behavior judgment algorithm for dangerous driving behavior detection, and the final output is the dangerous driving behavior judgment.
- the information of dangerous driving behavior is sent to the receiving terminal, which needs to be processed before sending;
- the external information includes start and end time, participants, video, etc., mainly involving the Web server mounted in the traffic monitoring LAN B.
- the web server receives the information output by the on-site workstation through the traffic monitoring LAN B, and expands the external information, and retrieves the location of the information and the corresponding video clip through the real-time/historical database to form a complete dangerous driving behavior information.
- the receiving terminal is then screened using the occurrence location information, and the complete accident information is sent to the terminal user.
- the technical process is shown in Figure 6.
- the method and system provided by the present invention can intelligently judge whether dangerous driving behavior occurs in road traffic by automatically identifying and processing the motion relationship between objects.
- the invention solves the problem of identifying dangerous driving behaviors that are likely to cause accidents. For this reason, the traffic control department or insurance company can obtain this information at the first time, which is convenient for them to carry out education or various restraint behaviors of car owners based on this information, so as to be able to correct
- the driver's driving habits can reduce the probability of dangerous driving behavior as much as possible, thereby greatly reducing the occurrence of road traffic accidents and achieving the purpose of safe travel.
- the detection of dangerous driving behavior is carried out. That is, the present invention realizes the detection and identification of dangerous driving behaviors by using the multi-target tracking technology based on data association and the method based on reasoning.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Traffic Control Systems (AREA)
Abstract
The present invention relates to the technical field of feature identification, and particularly relates to a detection method and system for dangerous driving behavior. The method comprises: acquiring a monitoring image sequence; for the monitoring image sequence, obtaining a parameter matrix of a moving target by means of a moving target detection algorithm based on deep learning and a target rapid tracking algorithm based on target matching and distribution; and determining whether the parameter matrix of the target is greater than a target threshold value, and if so, deeming the behavior as dangerous driving behavior, otherwise, deeming the behavior as normal driving behavior. By means of the present invention, the movement relationship between target objects can be automatically identified and processed, such that whether dangerous driving behavior occurs in road traffic can be intelligently determined; and dangerous driving behavior that is highly likely to cause an accident is identified, which is conducive to developing education or various constraint behaviors for a vehicle owner according to the information, such that the driving habits of a driver can be corrected, and the occurrence probability of dangerous driving behavior is reduced as much as possible, thereby greatly reducing the occurrence of road traffic accidents, and thus achieving the aim of safe travel.
Description
本发明涉及特征识别技术领域，具体涉及一种危险驾驶行为的检测方法以及系统。The invention relates to the technical field of feature recognition, in particular to a detection method and system for dangerous driving behavior.
我国每年的交通事故绝对数量是一个十分巨大的数字，造成了巨大的死亡人数和经济损失。而造成交通事故的一个很重要原因就是驾驶员的各种危险驾驶操作行为。The absolute number of traffic accidents in our country every year is a very huge number, resulting in huge death toll and economic losses. A very important cause of traffic accidents is the various dangerous driving behaviors of drivers.
如果道路驾驶员的驾驶行为能够得到有效识别和监管，则能够通过类似于违章处罚类的手段规范驾驶员的驾驶习惯，从而减少驾驶员不良驾驶行为发生的概率，进而减少交通事故潜在发生的可能，最后就能够减少城市交通事故发生的数量。因此，对道路驾驶员驾驶行为准确检测和识别就有着非常重要的实用价值。If the driving behavior of road drivers can be effectively identified and supervised, the driver’s driving habits can be regulated through means similar to violation penalties, thereby reducing the probability of drivers’ bad driving behaviors, thereby reducing the potential for traffic accidents , Finally, the number of urban traffic accidents can be reduced. Therefore, it has very important practical value to accurately detect and recognize the driving behavior of road drivers.
同时，现有危险驾驶行为识别技术仅能够针对单车识别一些简单的违章行为，针对车与车之间，车与人之间的危险行为是无法识别的。虽然，单车发生的违章行为是造成交通事故的潜在原因，但道路交通多目标直接交互而产生的危险驾驶行为才是最有可能造成事故的原因。为此，识别违章之外的危险驾驶行为就显得尤为重要，这样就可以对一些不违章，但可能直接造成事故的危险驾驶行为进行识别和处理，这样就能够大大提升道路交通事故发生的概率，大大减少人员生命和财产的损失。At the same time, the existing dangerous driving behavior recognition technology can only identify some simple violations for single vehicles, and cannot identify dangerous behaviors between vehicles and between vehicles. Although the violations of bicycles are the potential cause of traffic accidents, the most likely cause of accidents is the dangerous driving behavior caused by the direct interaction of multiple objects in road traffic. For this reason, it is particularly important to identify dangerous driving behaviors other than violations, so that some dangerous driving behaviors that do not violate the rules but may directly cause accidents can be identified and dealt with, which can greatly increase the probability of road traffic accidents. Greatly reduce the loss of human life and property.
发明内容Contents of the invention
本发明所要解决的技术问题是：提供一种危险驾驶行为智能检测处理系统，可以实现道路驾驶员危险驾驶行为的识别，进而通过干预手段，并最终降低城市道路交通事故发生的概率的危险驾驶行为的检测方法以及系统。The technical problem to be solved by the present invention is to provide an intelligent detection and processing system for dangerous driving behavior, which can realize the identification of dangerous driving behavior of road drivers, and then through intervention means, and finally reduce the probability of urban road traffic accidents. detection method and system.
为了解决上述技术问题，本发明采用的技术方案为：In order to solve the problems of the technologies described above, the technical solution adopted in the present invention is:
一种危险驾驶行为的检测方法，包括：A detection method for dangerous driving behavior, comprising:
获取监控图像序列；Obtain a sequence of surveillance images;
监控图像序列基于深度学习的运动目标检测算法、基于目标匹配和分配的目标快速跟踪算法得到运动目标的参数矩阵；The monitoring image sequence is based on the deep learning-based moving target detection algorithm, and the target fast tracking algorithm based on target matching and allocation to obtain the parameter matrix of the moving target;
判断目标的参数矩阵是否大于目标阈值，若是则视为危险驾驶行为，否则为正常驾驶行为。Determine whether the parameter matrix of the target is greater than the target threshold, if so, it is regarded as a dangerous driving behavior, otherwise it is a normal driving behavior.
优选地，所述参数矩阵包括目标ID、加速度矢量；Preferably, the parameter matrix includes target ID, acceleration vector;
判断运动目标的参数矩阵中目标ID的加速度矢量是否大于目标阈值进行，若是则视为危险驾驶行为，否则为正常驾驶行为。Judging whether the acceleration vector of the target ID in the parameter matrix of the moving target is greater than the target threshold, if it is, it is regarded as a dangerous driving behavior, otherwise it is a normal driving behavior.
优选地，所述基于深度学习的运动目标检测算法、基于目标匹配和分配的目标快速跟踪算法采用tracking by detection框架，其中detection负责运动目标的检测识别，tracking负责跟踪运动目标的帧间移动；Preferably, the moving target detection algorithm based on deep learning and the target fast tracking algorithm based on target matching and distribution adopt a tracking by detection framework, wherein detection is responsible for the detection and identification of moving targets, and tracking is responsible for tracking the inter-frame movement of moving targets;
所述基于目标匹配和分配的目标快速跟踪算法包括基于线性运动模型的卡尔曼预测模型、基于重叠度评价的目标匹配算法、基于匈牙利算法的目标分配算法。The target fast tracking algorithm based on target matching and allocation includes Kalman prediction model based on linear motion model, target matching algorithm based on overlap evaluation, and target allocation algorithm based on Hungarian algorithm.
优选地，所述基于线性运动模型的卡尔曼预测模型包括：Preferably, the Kalman prediction model based on the linear motion model includes:
预设运动目标在相邻两帧图像之间发生线性移动，进而预测该运动目标在下一帧图像中的位置；The preset moving target moves linearly between two adjacent frames of images, and then predicts the position of the moving target in the next frame of images;
通过输入状态向量和输出观测数据，进行最优估计；Perform optimal estimation by inputting state vectors and outputting observed data;
所述状态向量包括监控图像序列中的运动目标的位置参量、速度参量及检测框的尺寸参量、尺寸变化参量，以高斯白噪声为系统噪声和量测噪声，组建卡尔曼预测方程组后输出结果。The state vector includes the position parameter, velocity parameter and the size parameter of the detection frame and the size change parameter of the moving target in the monitoring image sequence, and Gaussian white noise is used as the system noise and measurement noise, and the Kalman prediction equations are set up to output the result .
优选地，所述基于重叠度评价的目标匹配算法包括：Preferably, the target matching algorithm based on overlap evaluation includes:
将同一运动目标在相邻两帧图像上的信息进行关联，描述运动目标的位置序列；Correlate the information of the same moving target on two adjacent frames of images, and describe the position sequence of the moving target;
通过重叠度评价基于深度学习的运动目标检测算法得出的结果和卡尔曼预测模型输出结果；Evaluate the results of the moving target detection algorithm based on deep learning and the output results of the Kalman prediction model through the degree of overlap;
所述重叠度评价的重叠度计算公式为：The overlapping degree calculation formula of the overlapping degree evaluation is:
式中c为candidatebound，即候选框；g为groundtruthbound，即原标记框；area表示面积；IOU为intersectionoverunion，即重叠度。In the formula, c is candidatebound, that is, the candidate frame; g is groundtruthbound, that is, the original mark frame; area represents the area; IOU is intersectionoverunion, that is, the degree of overlap.
优选地，所述基于匈牙利算法的目标分配算法包括：Preferably, the target allocation algorithm based on the Hungarian algorithm comprises:
通过基于重叠度评价的目标匹配算法计算得到相邻两帧图像中全部运动目标之间的一对多匹配，并从中建立最佳的一对一匹配；Calculate the one-to-many matching between all moving objects in two adjacent frames of images through the target matching algorithm based on the evaluation of overlap, and establish the best one-to-one matching;
将相邻两帧图像中的运动目标分配问题归纳为二部图问题，采用递归的方法，不断迭代寻找增广路径。The moving target allocation problem in two adjacent frames of images is summarized as a bipartite graph problem, and a recursive method is used to iteratively find the augmenting path.
优选地，所述基于深度学习的运动目标检测算法对监控图像序列中的目标进行分类和定位。Preferably, the moving target detection algorithm based on deep learning classifies and locates targets in the surveillance image sequence.
为了解决上述技术问题，本发明采用的另一技术方案为：In order to solve the above technical problems, another technical solution adopted by the present invention is:
一种危险驾驶行为的检测系统，包括：A detection system for dangerous driving behavior, comprising:
若干交通监控设备，产生视频流；A number of traffic monitoring devices, generating video streams;
交通监控局域网A，具有实时数据库服务器、现场工作站和监控工作站，实时数据库服务器对视频流进行存储，现场工作站对视频流采用权利要求1-7任意一项所述的危险驾驶行为的检测方法进行检测；监控工作站调取视频流并实时显示；The traffic monitoring local area network A has a real-time database server, an on-site workstation and a monitoring workstation, the real-time database server stores the video stream, and the on-site workstation uses the detection method for dangerous driving behavior described in any one of claims 1-7 to detect the video stream ;The monitoring workstation retrieves the video stream and displays it in real time;
交通监控局域网B，将交通监控局域网A的实时数据库服务器、现场工作站和监控工作站同时挂载到交通监控局域网B，通过Web服务器，实现对外信息的分发；以及The traffic monitoring local area network B mounts the real-time database server, on-site workstation and monitoring workstation of the traffic monitoring local area network A to the traffic monitoring local area network B at the same time, and realizes the distribution of external information through the Web server; and
接收终端，接收来自交通监控局域网B分发的对外信息。The receiving terminal receives the external information distributed from the traffic monitoring LAN B.
优选地，所述现场工作站对视频流采用权利要求1-7任意一项所述的危险驾驶行为的检测方法进行检测前先对视频流进行解码、进行运动目标的提取，并对运动目标进行分类。Preferably, the on-site workstation first decodes the video stream, extracts the moving object, and classifies the moving object before using the dangerous driving behavior detection method described in any one of claims 1-7 to detect the video stream .
优选地，所述对外信息包括起止时间、参与方、视频；Preferably, the external information includes start and end time, participants, and video;
所述交通监控局域网B还包括实时/历史数据库，所述实时/历史数据库将交通监控局域网A的实时数据库服务器上传的信息进行存储；The traffic monitoring local area network B also includes a real-time/historical database, and the real-time/historical database stores the information uploaded by the real-time database server of the traffic monitoring local area network A;
所述Web服务器通过交通监控局域网B接收现场工作站输出的信息，并对信息进行扩展，通过实时/历史数据库调取信息的发生地点和对应的视频片段，形成完整的危险驾驶行为信息；The Web server receives the information output by the field workstation through the traffic monitoring local area network B, and expands the information, and retrieves the place of occurrence of the information and the corresponding video clips through the real-time/historical database to form complete dangerous driving behavior information;
通过发生地点信息对接收终端进行筛选终端用户，并将包括完整事故信息的对外信息发送给终端用户对应的接收终端。The terminal user is screened by the receiving terminal based on the occurrence location information, and the external information including complete accident information is sent to the receiving terminal corresponding to the terminal user.
本发明的有益效果在于：可以通过自动识别和处理各目标物之间的运动关系，能够智能判断道路交通是否发生危险驾驶行为；解决了对极可能造成事故的危险驾驶行为进行识别，为此，交管部门或者保险公司能够第一时间获取此信息，便于其依据此信息开展车主的教育或者各项约束行为，从而能够纠正驾驶员的驾驶习惯，尽可能减少危险驾驶行为发生的概率，从而极大可能减少道路交通事故的发生，达到了安全出行的目的。The beneficial effect of the present invention is that it can intelligently judge whether dangerous driving behavior occurs in road traffic by automatically identifying and processing the motion relationship between objects; it solves the problem of identifying dangerous driving behavior that is likely to cause accidents. Traffic control departments or insurance companies can obtain this information at the first time, so that they can carry out education or various restraint behaviors of car owners based on this information, so as to correct the driving habits of drivers and reduce the probability of dangerous driving behaviors as much as possible, thus greatly It is possible to reduce the occurrence of road traffic accidents and achieve the purpose of safe travel.
图1为本发明具体实施方式的一种危险驾驶行为的检测方法的流程示意图；Fig. 1 is a schematic flow chart of a detection method for dangerous driving behavior according to a specific embodiment of the present invention;
图2为本发明具体实施方式的一种危险驾驶行为的检测方法的基于深度学习的运动目标检测算法、基于目标匹配和分配的目标快速跟踪算法流程示意图；2 is a schematic flowchart of a detection method for dangerous driving behavior based on deep learning and a fast-tracking algorithm based on target matching and distribution in a specific embodiment of the present invention;
图3为本发明具体实施方式的一种危险驾驶行为的检测方法的危险驾驶判定算法示意图；3 is a schematic diagram of a dangerous driving judgment algorithm of a dangerous driving behavior detection method according to a specific embodiment of the present invention;
图4为本发明具体实施方式的一种危险驾驶行为的检测系统的示意图；4 is a schematic diagram of a detection system for dangerous driving behavior according to a specific embodiment of the present invention;
图5为本发明具体实施方式的一种危险驾驶行为的检测系统的检测流程示意图；5 is a schematic diagram of a detection process of a dangerous driving behavior detection system according to a specific embodiment of the present invention;
图6为本发明具体实施方式的一种危险驾驶行为的检测方法的信息处理流程示意图。FIG. 6 is a schematic diagram of an information processing flow of a dangerous driving behavior detection method according to a specific embodiment of the present invention.
为详细说明本发明的技术内容、所实现目的及效果，以下结合实施方式并配合附图予以说明。In order to describe the technical content, achieved goals and effects of the present invention in detail, the following descriptions will be made in conjunction with the embodiments and accompanying drawings.
实施例一Embodiment one
参照图1，一种危险驾驶行为的检测方法，包括：Referring to Figure 1, a detection method for dangerous driving behavior, including:
通过对监控图像序列应用基于深度学习的运动目标检测算法、基于目标匹配和分配的目标快速跟踪算法得到目标的参数矩阵，其参数包含目标ID、加速度矢量等信息。其后，通过对目标参数矩阵中目标ID加速度矢量与目标阈值进行比对，从而判断是否发生危险驾驶行为。The parameter matrix of the target is obtained by applying the moving target detection algorithm based on deep learning and the target fast tracking algorithm based on target matching and allocation to the monitoring image sequence, and its parameters include target ID, acceleration vector and other information. Thereafter, by comparing the target ID acceleration vector in the target parameter matrix with the target threshold, it is judged whether dangerous driving behavior occurs.
所述危险驾驶行为的检测方法使用tracking by detection框架，其中detection负责运动目标的检测识别，tracking负责跟踪运动目标的帧间移动，其具体实现方式如图2所示。The detection method of the dangerous driving behavior uses a tracking by detection framework, wherein detection is responsible for the detection and recognition of moving objects, and tracking is responsible for tracking the inter-frame movement of moving objects, and its specific implementation is as shown in Figure 2.
(1)基于深度学习的目标检测网络(1) Target detection network based on deep learning
该网络对监控图像序列中的目标进行分类和定位。它的思想是通过多层神经网络对输入的信息进行衡量，通过大量数据不断计算学习，让机器学会如何去执行任务。该网络需要完成行人、非机动车、机动车等7类运动目标的分类和定位。根据需要，对网络的结构参数进行了相应的修改，将图像划分为7*7个网格、网格预测目标框设置为5个、输出种类设置为7类(行人、骑两轮车人、骑三轮车人、轿车、SUV、客车、卡车共七类)，这样通过该网络可以得到一个7*7*5*(7+5)维的运动目标检测结果。The network classifies and localizes objects in a sequence of surveillance images. Its idea is to measure the input information through a multi-layer neural network, and continuously calculate and learn through a large amount of data, so that the machine can learn how to perform tasks. The network needs to complete the classification and positioning of 7 types of moving objects such as pedestrians, non-motor vehicles, and motor vehicles. According to the needs, the structural parameters of the network were modified accordingly, the image was divided into 7*7 grids, the grid prediction target frame was set to 5, and the output types were set to 7 categories (pedestrians, people riding two-wheelers, There are seven types of tricycle riders, cars, SUVs, buses, and trucks), so that a 7*7*5*(7+5) dimensional moving target detection result can be obtained through the network.
(2)基于线性运动模型的卡尔曼预测模型(2) Kalman prediction model based on linear motion model
本发明假设运动目标在相邻两帧图像之间发生线性移动，进而预测该运动目标在下一帧图像中的位置。卡尔曼预测是一种利用线性系统状态方程，通过系统的输入状态向量和输出观测数据，对系统状态进行最优估计的算法。本发明涉及的状态向量由运动目标的位置参量(x，y)、速度参量(dx，dy)、及检测框的尺寸参量(w，h)、尺寸变化参量(dw，dh)组成，以高斯白噪声为系统噪声和量测噪声，组建卡尔曼预测方程组。The invention assumes that the moving object moves linearly between two adjacent frames of images, and then predicts the position of the moving object in the next frame of images. Kalman prediction is an algorithm that uses the linear system state equation to optimally estimate the system state through the input state vector and output observation data of the system. The state vector involved in the present invention is made up of the position parameter (x, y) of moving object, velocity parameter (dx, dy), and the size parameter (w, h) of detection frame, size variation parameter (dw, dh) of Gaussian The white noise is the system noise and the measurement noise, and the Kalman prediction equations are set up.
(3)基于重叠度评价的目标匹配算法(3) Target matching algorithm based on overlap evaluation
目标匹配算法可以将同一运动目标在相邻两帧图像上的信息进行关联，从而描述运动目标的位置序列。目标匹配算法的核心是用重叠度评价目标检测结果和卡尔曼预测模型输出结果。重叠度是目标检测中常使用的一个概念，是产生的候选框与原标记框的交叠率，计算如下式所示。The target matching algorithm can correlate the information of the same moving target on two adjacent frames of images, so as to describe the position sequence of the moving target. The core of the target matching algorithm is to use the overlapping degree to evaluate the target detection results and the output results of the Kalman prediction model. The degree of overlap is a concept often used in target detection. It is the overlap rate between the generated candidate frame and the original marked frame. The calculation is shown in the following formula.
式中c为candidate bound，即候选框；g为ground truth bound，即原标记框；area表示面积；IOU为intersection over union，即重叠度。当检测的精度很高，视频的帧率也很高时，重叠度的思想被引入目标跟踪。In the formula, c is the candidate bound, that is, the candidate frame; g is the ground truth bound, that is, the original marked frame; area represents the area; IOU is the intersection over union, that is, the degree of overlap. When the detection accuracy is high and the video frame rate is high, the idea of overlap is introduced into target tracking.
(4)基于匈牙利算法的目标分配算法(4) Target assignment algorithm based on Hungarian algorithm
因通过上述重叠度算法计算得到相邻两帧图像中全部运动目标之间的一对多匹配，目标分配算法可以从中建立最佳的一对一匹配。本发明将相邻两帧(k-1帧、k帧)图像中的运动目标分配问题归纳为二部图问题，即尽可能给k-1帧图像中最多的运动目标找到配对。通过使用递归的方法，不断迭代寻找增广路径，而每发现一条增广路径，就意味着一个更大匹配的出现。因此，本发明通过对二部图中的配对运动目标不断的拆分和重新组合，得到更大的匹配。Since the one-to-many matching between all moving objects in two adjacent frames of images is calculated through the above-mentioned overlapping degree algorithm, the object allocation algorithm can establish the best one-to-one matching therefrom. The present invention summarizes the problem of assigning moving objects in two adjacent frames (k-1 frame, k frame) into a bipartite graph problem, that is, finds pairs for the most moving objects in k-1 frame images as much as possible. By using a recursive method, iteratively searches for augmentation paths, and every discovery of an augmentation path means the emergence of a larger match. Therefore, the present invention obtains greater matching by continuously splitting and recombining the paired moving objects in the bipartite graph.
危险驾驶行为判定算法Judgment Algorithm for Dangerous Driving Behavior
本发明根据运动目标的ID、加速度矢量建立基于先验规则的危险驾驶行为判定模型，如图3所示。The present invention establishes a priori rule-based dangerous driving behavior judgment model according to the ID and acceleration vector of the moving object, as shown in FIG. 3 .
实施例二Embodiment two
参照图4，一种危险驾驶行为的检测系统，包括：Referring to Figure 4, a detection system for dangerous driving behavior, including:
多个交通监控设备通过OPCServer挂载到交通监控局域网A，其提供的交通监控视频流由实时数据库服务器、现场工作站和监控工作站同时处理。其中实时数据库服务器负责对局域网A内的视频流进行存储；监控工作站可以调取局域网A内的交通监控设备，并将视频实时显示；现场工作站对局域网A内的交通监控视频流进行实时处理，检测并提取危险驾驶行为。实时数据库服务器、现场工作站和监控工作站同时挂载到交通监控局域网B，通过Web服务器，可以实现对外信息的分发。系统的接收终端通过互联网接收Web服务器发送的消息。Multiple traffic monitoring devices are mounted to the traffic monitoring LAN A through the OPCServer, and the traffic monitoring video streams provided by it are processed simultaneously by the real-time database server, field workstation and monitoring workstation. Among them, the real-time database server is responsible for storing the video stream in LAN A; the monitoring workstation can call the traffic monitoring equipment in LAN A and display the video in real time; the on-site workstation can process and detect the traffic monitoring video stream in LAN A in real time And extract dangerous driving behavior. The real-time database server, on-site workstation and monitoring workstation are mounted to the traffic monitoring LAN B at the same time, and the distribution of external information can be realized through the Web server. The receiving terminal of the system receives the message sent by the Web server through the Internet.
危险驾驶行为技术主要涉及对交通监控局域网A中挂载的交通监控设备输出的视频进行处理，从中自动检测并提取危险驾驶行为数据，如图5所示。首先对交通监控视频流进行解码。然后使用深度神经网络YOLOv3进行运动目标 提取，运动目标的种类为行人、骑两轮车人、骑三轮车人、轿车、SUV、客车、卡车共七类。同时基于检测结果进行运动目标的跟踪，输出同一运动目标在一段时间内的参数矩阵，包括ID、位置和速度矢量。最后，将参数矩阵输入到危险驾驶行为判定算法进行危险驾驶行为检测，最终输出为危险驾驶行为判定。The dangerous driving behavior technology mainly involves processing the video output from the traffic monitoring equipment mounted in the traffic monitoring LAN A, and automatically detecting and extracting dangerous driving behavior data from it, as shown in Figure 5. Firstly, the traffic monitoring video stream is decoded. Then use the deep neural network YOLOv3 to extract moving objects. There are seven types of moving objects: pedestrians, people riding two-wheelers, people riding tricycles, cars, SUVs, buses, and trucks. At the same time, based on the detection results, the tracking of the moving target is carried out, and the parameter matrix of the same moving target in a period of time is output, including ID, position and velocity vector. Finally, the parameter matrix is input into the dangerous driving behavior judgment algorithm for dangerous driving behavior detection, and the final output is the dangerous driving behavior judgment.
将危险驾驶行为的信息(对外信息)发送到接收终端，发送前需要进行信息处理；对外信息包括起止时间、参与方、视频等，主要涉及交通监控局域网B中挂载的Web服务器。Web服务器通过交通监控局域网B接收现场工作站输出的信息，并对对外信息进行扩展，通过实时/历史数据库调取信息的发生地点和对应的视频片段，形成完整的危险驾驶行为信息。然后使用发生地点信息对接收终端进行筛选，并将完整事故信息发送给该终端用户。技术流程如图6所示。The information of dangerous driving behavior (external information) is sent to the receiving terminal, which needs to be processed before sending; the external information includes start and end time, participants, video, etc., mainly involving the Web server mounted in the traffic monitoring LAN B. The web server receives the information output by the on-site workstation through the traffic monitoring LAN B, and expands the external information, and retrieves the location of the information and the corresponding video clip through the real-time/historical database to form a complete dangerous driving behavior information. The receiving terminal is then screened using the occurrence location information, and the complete accident information is sent to the terminal user. The technical process is shown in Figure 6.
综上所述，本发明提供的方法和系统以通过自动识别和处理各目标物之间的运动关系，能够智能判断道路交通是否发生危险驾驶行为。本发明解决了对极可能造成事故的危险驾驶行为进行识别，为此，交管部门或者保险公司能够第一时间获取此信息，便于其依据此信息开展车主的教育或者各项约束行为，从而能够纠正驾驶员的驾驶习惯，尽可能减少危险驾驶行为发生的概率，从而极大可能减少道路交通事故的发生，达到了安全出行的目的。To sum up, the method and system provided by the present invention can intelligently judge whether dangerous driving behavior occurs in road traffic by automatically identifying and processing the motion relationship between objects. The invention solves the problem of identifying dangerous driving behaviors that are likely to cause accidents. For this reason, the traffic control department or insurance company can obtain this information at the first time, which is convenient for them to carry out education or various restraint behaviors of car owners based on this information, so as to be able to correct The driver's driving habits can reduce the probability of dangerous driving behavior as much as possible, thereby greatly reducing the occurrence of road traffic accidents and achieving the purpose of safe travel.
通过是针对多车开展多目标识别和跟踪，并对多车目标开展行驶轨迹、速度以及加速度等复杂行驶数据的计算和跟踪处理，并综合轨迹、速度以及加速度等数据分析相互之间的关系，并基于系统理论和危险驾驶行为判定规则进行危险驾驶行为的检测。即本发明使用了基于数据关联的多目标跟踪技术和基于推理的方法实现了危险驾驶行为的检测和识别。By carrying out multi-target recognition and tracking for multiple vehicles, and carrying out calculation and tracking processing of complex driving data such as driving trajectories, speeds and accelerations for multi-vehicle targets, and comprehensively analyzing the relationship between data such as trajectories, speeds and accelerations, And based on the system theory and the judgment rules of dangerous driving behavior, the detection of dangerous driving behavior is carried out. That is, the present invention realizes the detection and identification of dangerous driving behaviors by using the multi-target tracking technology based on data association and the method based on reasoning.
以上所述仅为本发明的实施例，并非因此限制本发明的专利范围，凡是利用本发明说明书及附图内容所作的等同变换，或直接或间接运用在相关的技术领域，均同理包括在本发明的专利保护范围内。The above description is only an embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent transformations made by using the description of the present invention and the contents of the accompanying drawings, or directly or indirectly used in related technical fields, are all included in the same principle. Within the scope of patent protection of the present invention.
Claims (10)
- 一种危险驾驶行为的检测方法，其特征在于，包括：A detection method for dangerous driving behavior, characterized in that it comprises:获取监控图像序列；Obtain a sequence of surveillance images;所述监控图像序列基于深度学习的运动目标检测算法、基于目标匹配和分配的目标快速跟踪算法得到运动目标的参数矩阵；The monitoring image sequence is based on a moving target detection algorithm based on deep learning, and a target fast tracking algorithm based on target matching and allocation to obtain a parameter matrix of a moving target;判断所述目标的参数矩阵是否大于目标阈值，若是则视为危险驾驶行为，否则为正常驾驶行为。It is judged whether the parameter matrix of the target is greater than the target threshold, if so, it is regarded as dangerous driving behavior, otherwise it is normal driving behavior.
- 根据权利要求1所述的危险驾驶行为的检测方法，其特征在于，所述参数矩阵包括目标ID、加速度矢量；The method for detecting dangerous driving behavior according to claim 1, wherein the parameter matrix includes a target ID and an acceleration vector;判断所述运动目标的参数矩阵中目标ID的加速度矢量是否大于目标阈值，若是则视为危险驾驶行为，否则为正常驾驶行为。Judging whether the acceleration vector of the target ID in the parameter matrix of the moving target is greater than the target threshold, if so, it is regarded as a dangerous driving behavior, otherwise it is a normal driving behavior.
- 根据权利要求1所述的危险驾驶行为的检测方法，其特征在于，所述基于深度学习的运动目标检测算法、基于目标匹配和分配的目标快速跟踪算法采用tracking by detection框架，其中detection负责运动目标的检测识别，tracking负责跟踪运动目标的帧间移动；The detection method of dangerous driving behavior according to claim 1, characterized in that, the moving target detection algorithm based on deep learning, the target fast tracking algorithm based on target matching and distribution adopt a tracking by detection framework, wherein detection is responsible for moving targets The detection and recognition of tracking is responsible for tracking the inter-frame movement of moving targets;所述基于目标匹配和分配的目标快速跟踪算法包括基于线性运动模型的卡尔曼预测模型、基于重叠度评价的目标匹配算法、基于匈牙利算法的目标分配算法。The target fast tracking algorithm based on target matching and allocation includes Kalman prediction model based on linear motion model, target matching algorithm based on overlap evaluation, and target allocation algorithm based on Hungarian algorithm.
- 根据权利要求3所述的危险驾驶行为的检测方法，其特征在于，所述基于线性运动模型的卡尔曼预测模型包括：The detection method of dangerous driving behavior according to claim 3, is characterized in that, described Kalman predictive model based on linear motion model comprises:预设运动目标在相邻两帧图像之间发生线性移动，进而预测该运动目标在下一帧图像中的位置；The preset moving target moves linearly between two adjacent frames of images, and then predicts the position of the moving target in the next frame of images;通过输入状态向量和输出观测数据，进行最优估计；Perform optimal estimation by inputting state vectors and outputting observed data;所述状态向量包括所述监控图像序列中的运动目标的位置参量、速度参量及检测框的尺寸参量、尺寸变化参量，以高斯白噪声为系统噪声和量测噪声，组建卡尔曼预测方程组后输出结果。The state vector includes the position parameter, the velocity parameter and the size parameter and the size change parameter of the detection frame of the moving target in the monitoring image sequence, with Gaussian white noise as the system noise and measurement noise, after the Kalman prediction equations are set up Output the result.
- 根据权利要求4所述的危险驾驶行为的检测方法，其特征在于，所述基于重叠度评价的目标匹配算法包括：The detection method for dangerous driving behavior according to claim 4, wherein the target matching algorithm based on overlapping evaluation comprises:将同一运动目标在相邻两帧图像上的信息进行关联，描述运动目标的位置 序列；Associate the information of the same moving target on two adjacent frames of images, and describe the position sequence of the moving target;通过重叠度评价基于深度学习的运动目标检测算法得出的结果和卡尔曼预测模型输出结果；Evaluate the results of the moving target detection algorithm based on deep learning and the output results of the Kalman prediction model through the degree of overlap;所述重叠度评价的重叠度计算公式为：The overlapping degree calculation formula of the overlapping degree evaluation is:
式中c为候选框；g为原标记框；area表示面积；IOU为重叠度。In the formula, c is the candidate box; g is the original marked box; area is the area; IOU is the degree of overlap. - 根据权利要求5所述的危险驾驶行为的检测方法，其特征在于，所述基于匈牙利算法的目标分配算法包括：The detection method of dangerous driving behavior according to claim 5, is characterized in that, the target distribution algorithm based on Hungarian algorithm comprises:通过基于重叠度评价的目标匹配算法计算得到相邻两帧图像中全部运动目标之间的一对多匹配，并从中建立最佳的一对一匹配；Calculate the one-to-many matching between all moving objects in two adjacent frames of images through the target matching algorithm based on the evaluation of overlap, and establish the best one-to-one matching;将相邻两帧图像中的运动目标分配问题归纳为二部图问题，采用递归的方法，不断迭代寻找增广路径。The moving target allocation problem in two adjacent frames of images is summarized as a bipartite graph problem, and a recursive method is used to iteratively find the augmenting path.
- 根据权利要求3所述的危险驾驶行为的检测方法，其特征在于，所述基于深度学习的运动目标检测算法对监控图像序列中的目标进行分类和定位。The detection method of dangerous driving behavior according to claim 3, characterized in that, the moving object detection algorithm based on deep learning classifies and locates the objects in the monitoring image sequence.
- 一种危险驾驶行为的检测系统，其特征在于，包括：A detection system for dangerous driving behavior, characterized in that it includes:若干交通监控设备，产生视频流；A number of traffic monitoring devices, generating video streams;交通监控局域网A，具有实时数据库服务器、现场工作站和监控工作站，所述实时数据库服务器对所述视频流进行存储，所述现场工作站对所述视频流采用权利要求1-7任意一项所述的危险驾驶行为的检测方法进行检测；所述监控工作站调取所述视频流并实时显示；The traffic monitoring local area network A has a real-time database server, an on-site workstation and a monitoring workstation, the real-time database server stores the video stream, and the on-site workstation adopts the method described in any one of claims 1-7 for the video stream. The detection method of dangerous driving behavior detects; The monitoring workstation calls the video stream and displays it in real time;交通监控局域网B，将所述交通监控局域网A的实时数据库服务器、现场工作站和监控工作站同时挂载到交通监控局域网B，通过Web服务器，实现对外信息的分发；以及The traffic monitoring local area network B mounts the real-time database server, field workstation and monitoring workstation of the traffic monitoring local area network A to the traffic monitoring local area network B simultaneously, and realizes the distribution of external information through the Web server; and接收终端，接收来自所述交通监控局域网B分发的对外信息。The receiving terminal receives the external information distributed from the traffic monitoring local area network B.
- 根据权利要求8所述的危险驾驶行为的检测系统，其特征在于，所述现场工作站对所述视频流采用权利要求1-7任意一项所述的危险驾驶行为的检测方法进行检测前先对所述视频流进行解码、进行运动目标的提取，并对运动目 标进行分类。The detection system for dangerous driving behavior according to claim 8, wherein the on-site workstation detects the video stream before using the detection method for dangerous driving behavior described in any one of claims 1-7. The video stream is decoded, the moving object is extracted, and the moving object is classified.
- 根据权利要求8所述的危险驾驶行为的检测系统，其特征在于，所述对外信息包括起止时间、参与方、视频；The detection system for dangerous driving behavior according to claim 8, wherein the external information includes start and end time, participants, and video;所述交通监控局域网B还包括实时/历史数据库，所述实时/历史数据库将所述交通监控局域网A的实时数据库服务器上传的信息进行存储；The traffic monitoring local area network B also includes a real-time/historical database, and the real-time/historical database stores the information uploaded by the real-time database server of the traffic monitoring local area network A;所述Web服务器通过所述交通监控局域网B接收现场工作站输出的信息，并对所述信息进行扩展，通过所述实时/历史数据库调取信息的发生地点和对应的视频片段，形成完整的危险驾驶行为信息；The web server receives the information output by the on-site workstation through the traffic monitoring local area network B, and expands the information, and retrieves the place of occurrence of the information and the corresponding video clip through the real-time/historical database to form a complete dangerous driving behavioral information;通过发生地点信息对所述接收终端进行筛选终端用户，并将包括完整事故信息的对外信息发送给所述终端用户对应的接收终端。The terminal user is screened by the receiving terminal based on the occurrence location information, and the external information including complete accident information is sent to the receiving terminal corresponding to the terminal user.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111246850.6A CN113989787A (en) | 2021-10-26 | 2021-10-26 | Detection method and system for dangerous driving behaviors |
| CN202111246850.6 | 2021-10-26 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2023071397A1 true WO2023071397A1 (en) | 2023-05-04 |
Family
ID=79741519
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2022/111853 WO2023071397A1 (en) | 2021-10-26 | 2022-08-11 | Detection method and system for dangerous driving behavior |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN113989787A (en) |
| WO (1) | WO2023071397A1 (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113989787A (en) * | 2021-10-26 | 2022-01-28 | 北京国家新能源汽车技术创新中心有限公司 | Detection method and system for dangerous driving behaviors |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108091142A (en) * | 2017-12-12 | 2018-05-29 | 公安部交通管理科学研究所 | For vehicle illegal activities Tracking Recognition under highway large scene and the method captured automatically |
| CN111931582A (en) * | 2020-07-13 | 2020-11-13 | 中国矿业大学 | Image processing-based highway traffic incident detection method |
| CN113129602A (en) * | 2019-12-31 | 2021-07-16 | 宇龙计算机通信科技(深圳)有限公司 | Vehicle state monitoring method and device, storage medium and electronic equipment |
| CN113989787A (en) * | 2021-10-26 | 2022-01-28 | 北京国家新能源汽车技术创新中心有限公司 | Detection method and system for dangerous driving behaviors |
-
2021
- 2021-10-26 CN CN202111246850.6A patent/CN113989787A/en active Pending
-
2022
- 2022-08-11 WO PCT/CN2022/111853 patent/WO2023071397A1/en unknown
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108091142A (en) * | 2017-12-12 | 2018-05-29 | 公安部交通管理科学研究所 | For vehicle illegal activities Tracking Recognition under highway large scene and the method captured automatically |
| CN113129602A (en) * | 2019-12-31 | 2021-07-16 | 宇龙计算机通信科技(深圳)有限公司 | Vehicle state monitoring method and device, storage medium and electronic equipment |
| CN111931582A (en) * | 2020-07-13 | 2020-11-13 | 中国矿业大学 | Image processing-based highway traffic incident detection method |
| CN113989787A (en) * | 2021-10-26 | 2022-01-28 | 北京国家新能源汽车技术创新中心有限公司 | Detection method and system for dangerous driving behaviors |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113989787A (en) | 2022-01-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20240149882A1 (en) | Multiple exposure event determination | |
| Wang et al. | A Vision‐Based Video Crash Detection Framework for Mixed Traffic Flow Environment Considering Low‐Visibility Condition | |
| CN110781768A (en) | Target object detection method and device, electronic device and medium | |
| WO2023071397A1 (en) | Detection method and system for dangerous driving behavior | |
| Li et al. | DBUS: Human driving behavior understanding system | |
| Naik et al. | Implementation of YOLOv4 algorithm for multiple object detection in image and video dataset using deep learning and artificial intelligence for urban traffic video surveillance application | |
| Nowosielski et al. | Automatic analysis of vehicle trajectory applied to visual surveillance | |
| Quinn et al. | Traffic flow monitoring in crowded cities | |
| Barbosa et al. | IoT based real-time traffic monitoring system using images sensors by sparse deep learning algorithm | |
| Sein et al. | Object Detection, Classification and Counting for Analysis of Visual Events | |
| CN117315624A (en) | Obstacle detection method, vehicle control method, device, apparatus, and storage medium | |
| Jiao | [Retracted] Intelligent Research Based on Deep Learning Recognition Method in Vehicle‐Road Cooperative Information Interaction System | |
| US20230048304A1 (en) | Environmentally aware prediction of human behaviors | |
| Batapati et al. | Video analysis for traffic anomaly detection using support vector machines | |
| Tschürtz et al. | System of systems safety analysis and evaluation in ZalaZONE | |
| CN114627400A (en) | Lane congestion detection method and device, electronic equipment and storage medium | |
| Bello et al. | Motorcycling-Net: A Segmentation Approach for Detecting Motorcycling Near Misses | |
| Byzkrovnyi et al. | Comparison of Potential Road Accident Detection Algorithms for Modern Machine Vision System | |
| CN108596146A (en) | Road multi-target classification method | |
| Ahmad et al. | Comparative study of dashcam-based vehicle incident detection techniques | |
| Dao et al. | MM-TrafficRisk: A Video-based Fleet Management Application for Traffic Risk Prediction, Prevention, and Querying | |
| Harini et al. | The Traffic Violation Detection System using YoloV7 | |
| Patil et al. | Sequential Neural Network Model with Spatial-Temporal Attention Mechanism for Robust Lane Detection Using Multi Continuous Image Frames | |
| Nadipour et al. | A deep-learning-based SIoV framework in vehicle detection and counting system for Intelligent traffic management | |
| Akib et al. | Integrated Traffic Violations Detection System for the Highways of Bangladesh |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22885297 Country of ref document: EP Kind code of ref document: A1 |