Search

CN-115424172-B - Traffic event detection method, device, equipment and storage medium

CN115424172BCN 115424172 BCN115424172 BCN 115424172BCN-115424172-B

Abstract

The invention discloses a traffic event detection method, a device, equipment and a storage medium, wherein video frames of each local view divided by a global view of a ball machine are sequentially acquired by controlling the ball machine according to preset polling parameters; the method comprises the steps of obtaining a global visual field, rearranging video frames of all local visual fields obtained by a dome camera to obtain a video sequence set of the global visual field, carrying out multidimensional convolution operation on each video sequence unit in the video sequence set to obtain input features of the global visual field at each moment, inputting the obtained input features into a convolution neural network to extract visual features at each moment, and inputting the obtained visual features into a circulation neural network according to time sequence to analyze so as to obtain occurrence probability of each event at each moment. Only one zoom rotary spherical camera is deployed, the global view needing to be controlled is split into a plurality of local views, the hardware cost and the deployment and maintenance cost are reduced, the characteristics of light weight, low cost and the like are achieved, and the precise high-precision distinguishing and managing capability of real-time traffic events is improved.

Inventors

  • LIN HUANKAI
  • WANG XIANGXUE
  • CHEN LIJUN
  • HONG SHUGUANG

Assignees

  • 高新兴科技集团股份有限公司

Dates

Publication Date
20260512
Application Date
20220825

Claims (9)

  1. 1. A method of traffic event detection, the method comprising: the method comprises the steps that a ball machine is controlled to sequentially acquire video frames of each local view divided by the global view of the ball machine according to preset polling parameters; Rearranging all video frames of the local vision field acquired by the dome camera to obtain a video sequence set of the global vision field; Performing multidimensional convolution operation on each video sequence unit in the video sequence set to obtain the input characteristics of the global visual field at each moment; Inputting the obtained input features into a convolutional neural network to extract visual features at each moment; Inputting the obtained visual features into a cyclic neural network according to a time sequence for analysis to obtain occurrence probability of each event at each moment; Rearranging all the video frames of the local view acquired by the dome camera to obtain a video sequence set of the global view, wherein the rearranging comprises the following steps: Acquiring each local field of view And sequentially arrange the video frames according to a fixed period order ; The video sequence of the global view obtained after rearranging the video frames acquired by all the local views is as follows: ; Rearranging video frames taken from all local views at the same time to obtain video sequence units covering the global view ; For all video sequence units acquired According to the time sequence arrangement, obtaining a video sequence set W of the global view; Wherein, the K represents that the current video frame is the kth complete period, k=0, 1,..s, S is used to represent the number of complete periods, i represents that the current video frame is the ith frame of the current complete period, i=0, 1,., N is used to represent the number of image frames per complete period, j represents that the current video frame is the image of the jth partial view, j=0, 1,..m, M is used to represent the number of partial views, S, M, N >0.
  2. 2. The traffic event detection method according to claim 1, wherein the polling parameter determination process specifically includes: Dividing the global view to be monitored into a plurality of local views according to a preset rule; determining a visual field parameter of each local visual field according to the visual angle of each local visual field and the installation position of the dome camera, wherein the visual field parameter comprises a visual field angle and a visual field focal length; and determining the polling parameters of the dome camera according to the polling sequence preset for all the local fields of view, the preset conversion period and the field parameters of each local field of view.
  3. 3. The traffic event detection method according to claim 1, wherein the performing a multidimensional convolution operation on each video sequence unit in the video sequence set to obtain an input feature of a global view at each moment specifically comprises: Acquiring each video sequence unit Containing video frames ; For all video frames Performing multidimensional convolution operation to obtain the input characteristics of the global visual field at each moment ; Wherein, the Where k represents the current video frame as the kth complete period, k=0, 1,..s, S is used to represent the number of complete periods, i represents the current video frame as the ith frame of the current complete period, i=0, 1,..n, N is used to represent the number of image frames per complete period, j represents the current video frame as the image of the jth partial view, j=0, 1..m, M is used to represent the number of partial views, S, M, N >0, d is the d-th moment, at the current moment, i and j are known, corresponding to the current moment d, q represents the current video frame as the video frame of the qth channel, q=0, 1..c, c is used to represent the number of channels each video frame contains, A convolution operation is represented and is performed, Indicating that the stitching operation is performed on x.
  4. 4. The traffic event detection method according to claim 1, wherein the inputting the obtained input features into the convolutional neural network extracts the visual features at each moment, specifically comprising: Extracting input characteristics of each moment according to convolutional neural network Visual characteristics of (a) ; Wherein, the Net is a deep learning network, and d is the d-th time.
  5. 5. The traffic event detection method according to claim 1, wherein the step of inputting the obtained visual features into the recurrent neural network in time sequence for analysis to obtain occurrence probability of each event at each time, specifically comprises: Extracting time sequence information in visual characteristics by adopting long-short-term memory recurrent neural network, and analyzing to obtain occurrence probability of various traffic events at the current moment ; Wherein, the D represents the time point d, Indicating the probability of occurrence of the x-th traffic event at time d, x=0, 1, p, p >0, Indicating the visual characteristics of the instant d, Indicating the visual characteristics at time d-1.
  6. 6. The traffic event detection method according to claim 1, characterized in that the method further comprises; judging the occurrence probability of each event at each moment; when the occurrence probability is not less than the event of the preset threshold probability, the occurrence time and the occurrence probability of the event are recorded, and alarm information is generated.
  7. 7. A traffic event detection apparatus for implementing the traffic event detection method according to any one of claims 1 to 6, the apparatus comprising: the image acquisition module is used for controlling the dome camera to sequentially acquire video frames of each local view divided by the global view of the dome camera according to preset polling parameters; The video sequence set acquisition module is used for rearranging video frames of all local fields of view acquired by the dome camera to obtain a video sequence set of the global field of view; the input characteristic calculation module is used for carrying out multidimensional convolution operation on each video sequence unit in the video sequence set to obtain the input characteristic of the global field of view at each moment; the visual feature calculation module is used for inputting the obtained input features into the convolutional neural network to extract the visual features at each moment; the probability calculation module is used for inputting the obtained visual characteristics into the cyclic neural network according to the time sequence for analysis to obtain the occurrence probability of each event at each moment; The video sequence set obtaining module is specifically configured to rearrange video frames of all local views obtained by the dome camera to obtain a video sequence set of the global view, and specifically includes: Acquiring each local field of view And sequentially arrange the video frames according to a fixed period order ; The video sequence of the global view obtained after rearranging the video frames acquired by all the local views is as follows: ; Rearranging video frames taken from all local views at the same time to obtain video sequence units covering the global view ; For all video sequence units acquired According to the time sequence arrangement, obtaining a video sequence set W of the global view; Wherein, the K represents that the current video frame is the kth complete period, k=0, 1,..s, S is used to represent the number of complete periods, i represents that the current video frame is the ith frame of the current complete period, i=0, 1,., N is used to represent the number of image frames per complete period, j represents that the current video frame is the image of the jth partial view, j=0, 1,..m, M is used to represent the number of partial views, S, M, N >0.
  8. 8. A terminal device comprising a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, the processor implementing the traffic event detection method according to any of claims 1 to 6 when the computer program is executed.
  9. 9. A computer readable storage medium, characterized in that the computer readable storage medium comprises a stored computer program, wherein the computer program, when run, controls a device in which the computer readable storage medium is located to perform the traffic event detection method according to any of claims 1 to 6.

Description

Traffic event detection method, device, equipment and storage medium Technical Field The present invention relates to the field of traffic event detection technologies, and in particular, to a traffic event detection method, device, equipment, and storage medium. Background Traffic events refer to illegal or abnormal events that occur on roads in real time and affect road traffic. The traffic system mainly comprises three events, namely a first event related to various traffic participants, such as vehicle reverse running, illegal parking, suspected accident, non-motor vehicle intrusion, pedestrian intrusion and the like, a second artificial or natural event affecting road traffic, such as spills, mist, smoke and the like, and a traffic state, such as vehicle creep, congestion and the like. The traffic event detection utilizes computing power resources to analyze and detect videos acquired by monitoring cameras deployed on roads in real time, and when the traffic event is detected, early warning information is sent out. In the prior art, a gun-ball linkage technology is generally adopted to detect traffic events, a gun camera looks panoramic, and after the traffic events are found, the gun camera is driven to focus on a target for tracking analysis. The scheme needs two video acquisition devices, the hardware cost is high, in the deployment process of double-camera linkage, the gun camera and the ball camera are required to be calibrated, accumulated errors are easy to introduce, the target focusing is likely to fail, detailed information cannot be extracted, the imaging resolution is high, and therefore the image transmission bandwidth and the calculation time consumption are high. Disclosure of Invention In order to solve the problems, the invention provides a traffic event detection method, a device, equipment and a storage medium, which have the characteristics of light weight, low cost and the like, and improve the fine high-precision discrimination and management capability of real-time traffic events. The embodiment of the invention provides a traffic event detection method, which comprises the following steps: the method comprises the steps that a ball machine is controlled to sequentially acquire video frames of each local view divided by the global view of the ball machine according to preset polling parameters; Rearranging all video frames of the local vision field acquired by the dome camera to obtain a video sequence set of the global vision field; Performing multidimensional convolution operation on each video sequence unit in the video sequence set to obtain the input characteristics of the global visual field at each moment; Inputting the obtained input features into a convolutional neural network to extract visual features at each moment; and inputting the obtained visual characteristics into a cyclic neural network according to a time sequence for analysis, and obtaining the occurrence probability of each event at each moment. Preferably, the polling parameter determining process specifically includes: Dividing the global view to be monitored into a plurality of local views according to a preset rule; determining a visual field parameter of each local visual field according to the visual angle of each local visual field and the installation position of the dome camera, wherein the visual field parameter comprises a visual field angle and a visual field focal length; and determining the polling parameters of the dome camera according to the polling sequence preset for all the local fields of view, the preset conversion period and the field parameters of each local field of view. As a preferred solution, the rearranging the video frames of all the local views acquired by the dome camera to obtain the video sequence set of the global view specifically includes: the video frames of each partial view R j are acquired and are sequentially arranged according to a fixed period sequence Global view video sequence obtained by rearranging video frames acquired from all local views ...... The method comprises the following steps: Rearranging video frames taken from all local views at the same time to obtain video sequence units covering the global view For all video sequence units acquiredAccording to the time sequence arrangement, obtaining a video sequence set W of the global view; Wherein, the K represents that the current video frame is the kth complete period, k=0, 1,..s, S is used to represent the number of complete periods, i represents that the current video frame is the ith frame of the current complete period, i=0, 1,..n, N is used to represent the number of image frames per complete period, j represents that the current video frame is the image of the jth partial view, j=0, 1..m, M is used to represent the number of partial views, S, M, N >0. Preferably, the multidimensional convolution operation is performed on each video sequence unit in the video sequence set to obtain an input feature