WO2017177902A1 - 视频录制方法、服务器、系统及存储介质 - Google Patents

视频录制方法、服务器、系统及存储介质 Download PDF

Info

Publication number
WO2017177902A1
WO2017177902A1 PCT/CN2017/080113 CN2017080113W WO2017177902A1 WO 2017177902 A1 WO2017177902 A1 WO 2017177902A1 CN 2017080113 W CN2017080113 W CN 2017080113W WO 2017177902 A1 WO2017177902 A1 WO 2017177902A1
Authority
WO
WIPO (PCT)
Prior art keywords
region
sub
image
area
pixel
Prior art date
Application number
PCT/CN2017/080113
Other languages
English (en)
French (fr)
Inventor
王健宗
夏磊豪
马进
刘铭
肖京
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Priority to SG11201800364YA priority Critical patent/SG11201800364YA/en
Priority to US15/737,323 priority patent/US10349003B2/en
Priority to KR1020187019521A priority patent/KR102155182B1/ko
Priority to EP17781878.8A priority patent/EP3445044B1/en
Priority to AU2017250159A priority patent/AU2017250159B2/en
Priority to JP2018524835A priority patent/JP6425856B1/ja
Publication of WO2017177902A1 publication Critical patent/WO2017177902A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00—Details of television systems
    • H04N5/76—Television signal recording
    • H04N5/91—Television signal processing therefor
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/254—Analysis of motion involving subtraction of images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00—Details of television systems
    • H04N5/76—Television signal recording
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00—Details of television systems
    • H04N5/76—Television signal recording
    • H04N5/765—Interface circuits between an apparatus for recording and another apparatus
    • H04N5/77—Interface circuits between an apparatus for recording and another apparatus between a recording apparatus and a television camera
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N7/00—Television systems
    • H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
    • H04N7/181—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast for receiving images from a plurality of remote sources
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G06T2207/10021—Stereoscopic video; Stereoscopic image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10081—Computed x-ray tomography [CT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20021—Dividing image into blocks, subimages or windows
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30232—Surveillance
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/44—Event detection

Definitions

  • the present application relates to the field of video processing technologies, and in particular, to a video recording method, a server, a system, and a storage medium.
  • the monitoring system will continuously record the image 24 hours a day, so the image recording will be performed when the car is not repaired, and a large number of still video frames will be generated. Retaining a large number of such static video frames can result in wasted storage space and network bandwidth.
  • key information can be found from long static video frames, wasting time and effort, and may even miss key frames.
  • the existing video recording method monitors the panorama, and the recording action is triggered only when an action occurs, and such a function can alleviate the above problem to some extent.
  • the problem is that unrelated actions can also trigger the recording action. For example, when there is a pedestrian passing 5 meters away from the vehicle to be repaired, although there is no relationship with the vehicle to be repaired, the video recording will be performed because of the action triggering, and information redundancy will be caused.
  • the present application provides a video recording method, server, system, and storage medium that can reduce the recording of unnecessary video frames.
  • the video recording methods provided by the application include:
  • the surveillance camera is controlled to start video recording from the currently extracted second image.
  • the server provided by the application includes a storage device and a processor, wherein:
  • the storage device is configured to store a video recording system
  • the processor is configured to invoke and execute the video recording system to perform the following steps:
  • the surveillance camera is controlled to start video recording from the currently extracted second image.
  • the video recording system provided by the present application includes:
  • a first image acquisition module configured to extract a first image captured by a surveillance camera every first preset time period
  • a modeling module configured to perform area detection on the extracted first image by using a pre-established model to extract an area of interest including a part or all parts of the target;
  • a screening module configured to perform a motion region screening on the region of interest by using an analysis rule to filter out a target region
  • a segmentation module configured to segment the selected target regions according to a segmentation rule, to divide each target region into a plurality of sub-regions
  • a second image acquisition module configured to extract a second image captured by the surveillance camera every second preset time
  • a motion detection module configured to compare an image block in each sub-region of the second image with an image block of a second image that is previously extracted in the same sub-region to determine whether a motion event occurs in each sub-region ;
  • the video recording module is configured to control the surveillance camera to perform video recording from the currently extracted second image when a motion event occurs in a certain sub-area.
  • the present application provides a non-volatile storage medium having computer readable instructions executable by one or more processors to perform the following steps:
  • the surveillance camera is controlled to start video recording from the currently extracted second image.
  • the video recording method and the server, the system and the storage medium applicable to the method can reduce unnecessary video frame recording and reduce waste of storage space and network bandwidth.
  • FIG. 1 is a schematic diagram of a server application environment of a first preferred embodiment of a video recording system of the present application.
  • FIG. 2 is a schematic diagram of a terminal application environment of a second preferred embodiment of the video recording system of the present application.
  • FIG. 3 is a functional block diagram of a preferred embodiment of the video recording system of the present application.
  • FIG. 4 is a flow chart of a method implementation of a preferred embodiment of the video recording method of the present application.
  • FIG. 5 is a detailed implementation flowchart of determining whether a motion event has occurred in each sub-area in the preferred embodiment of the video recording method of FIG. 4.
  • FIG. 1 it is a schematic diagram of a server application environment of a first preferred embodiment of the video recording system of the present application.
  • the video recording system 10 can be installed and run in a server.
  • the server may be a monitoring server 1.
  • the monitoring server 1 can be communicatively coupled to one or more surveillance cameras 3 installed in a monitoring location 2 via a communication module (not shown).
  • the monitoring place 2 can be a school, a kindergarten, a shopping mall, a hospital, a park, a city. Places such as squares and underground pedestrian passages can also be special areas where installation and monitoring are required, such as homes, small supermarkets, and ATM (Automatic Teller Machine) machines. In this embodiment, the monitoring place 2 is an automobile repair shop, such as a 4S shop.
  • the surveillance camera 3 can be an analog camera.
  • the analog camera can convert the analog video signal generated by the video capture device into a digital signal through a specific video capture card, and then transmit and store it in the monitoring server 1.
  • the surveillance camera 3 is a webcam. After the network camera is fixed, the network cable is connected to the router, and is connected to the monitoring server 1 through a router to perform video output by the monitoring server 1.
  • the monitoring server 1 may include a processor and a storage device (not shown).
  • the processor is a Core Unit and a Control Unit for interpreting computer instructions and processing data in computer software.
  • the storage device stores a database, an operating system, and the video recording system 10 described above.
  • the storage device includes an internal memory and a non-volatile storage medium; the video recording system, an operating system, and a database are stored on a non-volatile storage medium; and the internal memory is an operating system, a database, and a video recording system. 10 provides a cached runtime environment.
  • the video recording system 10 includes at least one computer-executable program instruction code, which can be executed by a processor to implement the following operations.
  • the pre-established model is a Convolutional Neural Network (CNN) model.
  • CNN Convolutional Neural Network
  • the model generation step includes:
  • the CNN model of the preset model structure is trained using a preset number of images after the area in which the vehicle is located to generate a CNN model that identifies the region of interest in the image.
  • the purpose of the training is to optimize the values of the weights within the CNN network so that the network model as a whole can actually be better applied to the identification of the region of interest.
  • the network model has a total of seven layers, five convolutional layers, one downsampled layer, and one fully connected layer.
  • the convolutional layer is formed by a feature map constructed by a plurality of feature vectors, and the function of the feature map is to extract key features by using a convolution filter.
  • the function of the downsampling layer is to remove the feature points of repeated expression and reduce the number of feature extractions by sampling method, thereby improving the efficiency of data communication between network layers.
  • the available sampling methods include maximum sampling method, mean sampling method and random sampling method.
  • the role of the fully connected layer is to connect the previous convolutional layer with downsampling and calculate the weight matrix for subsequent actual classification.
  • each layer undergoes two processes of forward iteration and backward iteration. Each iteration generates a probability distribution. The probability distributions after multiple iterations are superimposed, and the system selects the category with the largest value in the probability distribution. As a final classification result.
  • the analysis rule is: analyzing whether the extracted interest area is within a preset pixel area, for example, the preset pixel area range includes an abscissa area range and an ordinate area range, wherein the abscissa area
  • the range is (X1, X2), the ordinate range is (Y1, Y2), the X1 represents the X1 column pixel, the X2 represents the X2 column pixel, and X1 is smaller than X2.
  • Y1 represents the Y1th row of pixels, Y2 represents the Y2th row of pixels, and Y1 is smaller than Y2; if the extracted region of interest is within the preset pixel region, it is confirmed that the region of interest is The target area.
  • the principle of the analysis rule is that the monitoring of the repair shop is generally directed to a repair station to ensure that the vehicle occupies the main area of the lens, that is, the middle area. Therefore, the preset pixel area range should cover the main area of the lens as much as possible; The range should not be too large to prevent multiple areas of interest from falling into it; the range should not be too small to prevent the target area from falling into it; the range of the abscissa and the range of the ordinate can be verified manually. If it is too large, it will be reduced. If it is too small, it will be adjusted.
  • the segmentation rule is: a uniform segmentation mode is adopted, that is, the size and shape of the segmented sub-regions are consistent, and the target region is divided into sub-regions; and the target region is divided into N*N sub-regions, wherein N is a positive integer greater than 1, for example, 8*8.
  • N is a positive integer greater than 1, for example, 8*8.
  • the frame can be saved, and it is not necessary to continue to detect other parts. For example, in one example, taking 8*8 sub-regions as an example, if an action is detected on the first sub-area, there is no need to detect the remaining 63 sub-regions, thereby improving efficiency. 64 times.
  • the step of determining whether each of the sub-areas has a motion event comprises: placing the currently extracted second image in a pixel value of each pixel of the image block in each sub-area, and being in the same sub-sample as the second image extracted last time.
  • the image block of the region is compared with the pixel value of the pixel; the total difference corresponding to the image block in each sub-region is summed, and the calculated sum is divided by the number of pixels of the image block to obtain each sub- The average value of the difference corresponding to the image block in the area; and if the average value of the difference corresponding to the image block in the sub-area is greater than a preset threshold, it is determined that a motion event has occurred in the sub-area.
  • the video recording system 10 can also be installed and run in any one of the terminal devices, such as the mobile terminal 4 shown in FIG.
  • the mobile terminal 4 can be any electronic device with certain data processing functions, such as a smart phone, a tablet computer, a notebook computer, a wearable watch, wearable glasses, and the like.
  • the terminal device 2 also includes a processor and a storage device (not shown), the video recording system 10 including at least one computer-executable program instruction code stored in the storage device of the terminal device 2; The operations described in the first embodiment are implemented under the execution of the processor of the terminal device 2.
  • FIG. 1 and FIG. 2 are only block diagrams of partial structures related to the solution of the present application, and do not constitute a limitation of the server or the terminal device of the solution of the present application. Specifically, the electronic device More or fewer components than those shown in the figures may be included, or some components may be combined, or have different component arrangements.
  • non-volatile storage medium in the foregoing embodiment may be a non-volatile storage medium such as a magnetic disk, an optical disk, a read-only memory (ROM), or a random storage memory (Random Access). Memory, RAM), etc.
  • a non-volatile storage medium such as a magnetic disk, an optical disk, a read-only memory (ROM), or a random storage memory (Random Access). Memory, RAM), etc.
  • the storage device can be built in or externally connected to the monitoring server 1 or the terminal device 2.
  • FIG. 3 it is a functional block diagram of a preferred embodiment of the video recording system of the present invention.
  • the program code of the video recording system 10 can be divided into a plurality of functional modules according to different functions thereof.
  • the video recording system 10 may include a first image acquisition module 100, a modeling module 101, a screening module 102, a segmentation module 103, a second image acquisition module 104, a motion detection module 105, and a video. Recording module 106.
  • the first image acquisition module 100 is configured to extract a first image captured by the surveillance camera 3 every first preset time period, such as every 5 minutes.
  • the modeling module 101 is configured to perform area detection on the extracted first image by using a pre-established model to extract an interest area including a target such as a part or all parts of the vehicle.
  • the pre-established model is a convolutional neural network (Convolutional Neural Network, CNN) model.
  • CNN Convolutional Neural Network
  • the preset type model generating step includes:
  • the area where the vehicle is located is marked in each of the collected photos, wherein in the process of labeling, the position of the vehicle can be marked with a rectangular frame and the label is given. This process can be carried out in the form of crowdsourcing or manual labeling within the company, and the area where the marked vehicle is located is the area of interest.
  • the CNN model of the preset model structure is trained using a preset number of images after the area in which the vehicle is located to generate a CNN model that identifies the region of interest in the image.
  • the purpose of the training is to optimize the values of the weights within the CNN network so that the network model as a whole can actually be better applied to the identification of the region of interest.
  • the network model has a total of seven layers, five convolutional layers, one downsampled layer, and one fully connected layer.
  • the convolutional layer is formed by a feature map constructed by a plurality of feature vectors, and the function of the feature map is to extract key features by using a convolution filter.
  • the function of the downsampling layer is to remove the feature points of repeated expression and reduce the number of feature extractions by sampling method, thereby improving the efficiency of data communication between network layers.
  • the available sampling methods include maximum sampling method, mean sampling method and random sampling method.
  • the role of the fully connected layer is to connect the previous convolutional layer with downsampling and calculate the weight matrix for subsequent actual classification. After the image enters the model, each layer undergoes two processes of forward iteration and backward iteration. Each iteration generates a probability distribution. The probability distributions after multiple iterations are superimposed, and the system selects the category with the largest value in the probability distribution. As a final classification result.
  • the screening module 102 is configured to perform motion region screening on the region of interest by using an analysis rule to filter out the target region.
  • the analysis rule is: analyzing whether the extracted interest area is within a preset pixel area, for example, the preset pixel area range includes an abscissa area range and an ordinate area range, wherein the abscissa area The range is (X1, X2), the ordinate area range is (Y1, Y2); if the extracted interest area is within the preset pixel area range, it is confirmed that the interest area is the target area.
  • the principle of the analysis rule is that the monitoring of the repair shop is generally directed to a repair station to ensure that the vehicle occupies the main area of the lens, that is, the middle area.
  • the preset pixel area range should cover the main area of the lens as much as possible;
  • the range should not be too large to prevent multiple areas of interest from falling into it; the range should not be too small to prevent the target area from falling into it; the range of the abscissa and the range of the ordinate can be verified manually. If it is too large, it will be reduced. If it is too small, it will be adjusted.
  • the segmentation module 103 is configured to segment the selected target regions according to a segmentation rule to divide each target region into a plurality of sub-regions.
  • the segmentation rule is: a uniform segmentation method, that is, a segmented sub-region surface
  • the product size and shape are consistent, and the target area is divided into sub-areas; the target area is divided into N*N sub-areas, where N is a positive integer greater than 1, for example, 8*8.
  • N is a positive integer greater than 1, for example, 8*8.
  • N*N sub-areas for motion detection instead of for the overall target.
  • the first is accuracy. If only the target value is compared for the target value, the smaller motion may be averaged by other static parts. If it is dropped, it is impossible to detect such a subtle action; the second is efficiency, and the possible action only occurs in a certain area.
  • the frame can be saved, and it is not necessary to continue to detect other parts. For example, in one example, taking 8*8 sub-regions as an example, if an action is detected on the first sub-area, it is not necessary to detect the remaining 63 sub-regions, thereby improving the efficiency by 64 times.
  • the second image obtaining module 104 is configured to extract the second image captured by the surveillance camera 3 every second preset time, such as 0.5 seconds.
  • the motion detection module 105 is configured to compare an image block in each sub-region in the second image with an image block in which the second image extracted in the same sub-region is in the same sub-region to determine whether each sub-region has occurred. Sports events.
  • the step of determining whether each of the sub-areas has a motion event comprises: the pixel value of each pixel of the image block in which the currently extracted second image is in one of the sub-areas is the same as the previous image extracted
  • the image block of the sub-region corresponds to the pixel value of the pixel, and the difference is corresponding; all the differences corresponding to the image block in the sub-region are summed, and the calculated sum is divided by the number of pixels of the image block to obtain the The average value of the difference corresponding to the image block in the sub-area; and if the average value of the difference corresponding to the image block in the sub-area is greater than a preset threshold, it is determined that a motion event occurs in the sub-area.
  • the video recording module 106 is configured to control the surveillance camera 3 to perform video recording from the currently extracted second image when a motion event occurs in a certain sub-area.
  • FIG. 4 it is a flowchart of a method implementation of a preferred embodiment of the video recording method of the present invention.
  • the video recording method in this embodiment is not limited to the steps shown in the flowchart. In addition, in the steps shown in the flowchart, some steps may be omitted, and the order between the steps may be changed.
  • step S10 the first image acquisition module 100 extracts a first image captured by the surveillance camera 3 every first preset time period, such as every 5 minutes.
  • step S11 the modeling module 101 performs area detection on the extracted first image by using a pre-established model to extract an area of interest including a target object, such as a part of the vehicle or all parts.
  • the pre-established model is a Convolutional Neural Network (CNN) model.
  • CNN Convolutional Neural Network
  • the preset type model generating step includes:
  • the area where the vehicle is located is marked in each of the collected photos, wherein in the process of labeling, the position of the vehicle can be marked with a rectangular frame and the label is given. This process can be carried out in the form of crowdsourcing or manual labeling within the company, and the area where the marked vehicle is located is the area of interest.
  • the CNN model of the preset model structure is trained using a preset number of images after the area in which the vehicle is located to generate a CNN model that identifies the region of interest in the image.
  • the purpose of the training is to optimize the values of the weights within the CNN network so that the network model as a whole can actually be better applied to the identification of the region of interest.
  • the network model has a total of seven layers, five convolutional layers, one downsampled layer, and one fully connected layer.
  • the convolutional layer is formed by a feature map constructed by a plurality of feature vectors, and the function of the feature map is to extract key features by using a convolution filter.
  • the function of the downsampling layer is to remove the feature points of repeated expression and reduce the number of feature extractions by sampling method, thereby improving the efficiency of data communication between network layers.
  • the available sampling methods include maximum sampling method, mean sampling method and random sampling method.
  • the role of the fully connected layer is to connect the previous convolutional layer with downsampling and calculate the weight matrix for subsequent actual classification. After the image enters the model, each layer undergoes two processes of forward iteration and backward iteration. Each iteration generates a probability distribution. The probability distributions after multiple iterations are superimposed, and the system selects the category with the largest value in the probability distribution. As a final classification result.
  • step S12 the screening module 102 performs motion region screening on the region of interest by using an analysis rule to filter out the target region.
  • the analysis rule is: analyzing whether the extracted interest area is within a preset pixel area, for example, the preset pixel area range includes an abscissa area range and an ordinate area range, wherein the abscissa area The range is (X1, X2), the ordinate area range is (Y1, Y2); if the extracted interest area is within the preset pixel area range, it is confirmed that the interest area is the target area.
  • the principle of the analysis rule is that the monitoring of the repair shop is generally directed to a repair station to ensure that the vehicle occupies the main area of the lens, that is, the middle area.
  • the preset pixel area range should cover the main area of the lens as much as possible;
  • the range should not be too large to prevent multiple areas of interest from falling into it; the range should not be too small to prevent the target area from falling into it; the range of the abscissa and the range of the ordinate can be verified manually. If it is too large, it will be reduced. If it is too small, it will be adjusted.
  • step S13 the screening module 102 determines whether at least one target area is selected. When no target area is selected, return to step 10 above to re-execute the extraction of the first image. When the target area is screened, the following step S14 is performed.
  • step S14 the segmentation module 103 divides the selected target regions according to a segmentation rule to divide each target region into a plurality of sub-regions.
  • the segmentation rule is: a uniform segmentation mode is adopted, that is, the size and shape of the segmented sub-regions are consistent, and the target region is divided into sub-regions; and the target region is divided into N*N sub-regions, wherein N is a positive integer greater than 1, for example, 8*8.
  • N is a positive integer greater than 1, for example, 8*8.
  • the frame can be saved, and it is not necessary to continue to detect other parts. For example, in one example, taking 8*8 sub-regions as an example, if an action is detected on the first sub-area, it is not necessary to detect the remaining 63 sub-regions, thereby improving the efficiency by 64 times.
  • step S15 the second image acquisition module 104 extracts the second image captured by the surveillance camera 3 every second preset time, such as 0.5 seconds.
  • Step S16 the motion detection module 105 compares the image block in each sub-region of the second image with the image block of the second image extracted in the same sub-region to determine whether each sub-region has undergone motion. event.
  • the detailed implementation flow of step S16 is as described in the following FIG.
  • step S17 the motion detection module 105 determines whether a motion event has occurred in each sub-area. When no motion event has occurred in any of the sub-areas, the process returns to the above-described step S15. When a motion event has occurred in any of the sub-areas, the following step S18 is performed.
  • step S18 the video recording module 106 controls the surveillance camera 3 to perform video recording from the currently extracted second image.
  • step S16 in FIG. 4 it is a detailed implementation flowchart of step S16 in FIG. 4, that is, whether or not a motion event has occurred in each sub-area.
  • the video recording method in this embodiment is not limited to the steps shown in the flowchart. In addition, in the steps shown in the flowchart, some steps may be omitted, and the order between the steps may be changed.
  • step S160 the motion detection module 105 acquires pixel values of respective pixel points of the image block in which the currently extracted second image is in one of the sub-regions.
  • Step S161 the motion detection module 105 sets the currently extracted second image to the pixel value of each pixel of the image block in the sub-region and the pixel of the corresponding pixel of the image block of the same sub-region in the same sub-region.
  • the value is the difference.
  • Step S162 the motion detection module 105 sums all the differences corresponding to the image blocks in the sub-area, and divides the calculated sum by the number of pixels of the image block to obtain an image in the sub-area. The average value of the difference corresponding to the block.
  • step S163 the motion detection module 105 determines whether the average value of the difference corresponding to the image block of the sub-area is greater than a preset threshold. If the average value of the difference corresponding to the image block of the sub-area is greater than the preset threshold, step S164 is performed. Otherwise, when the average value of the difference corresponding to the image block of the sub-area is less than the preset threshold, step S165 is performed.
  • Step S164 the motion detection module 105 determines that a motion event has occurred in the sub-area.
  • Step S165 the motion detection module 105 determines that no motion event occurs in the sub-area, and returns to the foregoing step S160, and the motion detection module 105 acquires the current extracted The pixel value of each pixel of the image block in the next sub-area.
  • the storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)
  • Closed-Circuit Television Systems (AREA)
  • Television Signal Processing For Recording (AREA)

Abstract

一种视频录制方法，包括：每隔预设时间段抽取一监控摄像头所捕获的第一图像；对第一图像进行区域检测，以提取出包含目标物部分部位或者全部部位的兴趣区域；对所述兴趣区域执行运动区域筛选，以筛选出目标区域；对目标区域进行分割，以将每个目标区域分割成多个子区域；每隔预设时间抽取所述监控摄像头所捕获的第二图像；将第二图像中处于每个子区域中的图像块与前一次抽取的第二图像处于相同子区域的图像块进行比较，以判断各个子区域是否发生了运动事件；及当某一个子区域发生了运动事件时，控制所述监控摄像头从当前抽取的第二图像开始进行视频录制。提供一种适用于上述方法的服务器、系统及存储介质。

Description

视频录制方法、服务器、系统及存储介质
本申请要求于2016年4月14日提交中国专利局、申请号为201610234956.7，发明名称为“视频录制方法及服务器”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及视频处理技术领域，特别是一种视频录制方法、服务器、系统及存储介质。
背景技术
汽车在修理厂进行修理的过程中，监控系统会全天候不间断地进行影像录制，因此在没有进行汽车修理时也会进行影像录制，于是会产生大量的静态视频帧。保留大量这样的静态视频帧会造成存储空间和网络带宽的浪费。此外，在影像中查看检索关键信息时，要从长时间的静态视频帧中才能找到关键信息，浪费时间和精力，甚至可能会错过关键帧。
目前已有的视频录制方法会监控全景，只有当有动作发生时才会触发录制动作，这样的功能可以从一定程度上缓解上述问题。但存在的问题是不相关的动作也会触发录制动作。例如，在待修理车辆5米外有一个行人走过时，尽管和待修理车辆没有任何关系，但因为有动作触发，也会进行影像录制，还是会造成信息冗余。
发明内容
鉴于以上内容，本申请提供一种视频录制方法、服务器、系统及存储介质，其可以减少不必要的视频帧的录制。
本申请所提供的视频录制方法包括：
每隔第一预设时间段抽取一监控摄像头所捕获的第一图像；
利用一预先建立的模型，对抽取的第一图像进行区域检测，以提取出包含目标物部分部位或者全部部位的兴趣区域；
利用一分析规则，对所述兴趣区域执行运动区域筛选，以筛选出目标区域；
按照一分割规则，对所筛选出来的目标区域进行分割，以将每个目标区域分割成多个子区域；
每隔第二预设时间抽取所述监控摄像头所捕获的第二图像；
将所述第二图像中处于每个子区域中的图像块与前一次抽取的第二图像处于相同子区域的图像块进行比较，以判断各个子区域是否发生了运动事件；及
当某一个子区域发生了运动事件时，控制所述监控摄像头从当前抽取的第二图像开始进行视频录制。
本申请所提供的服务器包括存储设备以及处理器，其中：
所述存储设备，用于存储一个视频录制系统；
所述处理器，用于调用并执行所述视频录制系统，以执行如下步骤：
每隔第一预设时间段抽取一监控摄像头所捕获的第一图像；
利用一预先建立的模型，对抽取的第一图像进行区域检测，以提取出包含目标物部分部位或者全部部位的兴趣区域；
利用一分析规则，对所述兴趣区域执行运动区域筛选，以筛选出目标区域；
按照一分割规则，对所筛选出来的目标区域进行分割，以将每个目标区域分割成多个子区域；
每隔第二预设时间抽取所述监控摄像头所捕获的第二图像；
将所述第二图像中处于每个子区域中的图像块与前一次抽取的第二图像处于相同子区域的图像块进行比较，以判断各个子区域是否发生了运动事件；及
当某一个子区域发生了运动事件时，控制所述监控摄像头从当前抽取的第二图像开始进行视频录制。
本申请提供的视频录制系统包括：
第一图像获取模块，用于每隔第一预设时间段抽取一监控摄像头所捕获的第一图像；
建模模块，用于利用一预先建立的模型，对抽取的第一图像进行区域检测，以提取出包含目标物部分部位或者全部部位的兴趣区域；
筛选模块，用于利用一分析规则，对所述兴趣区域执行运动区域筛选，以筛选出目标区域；
分割模块，用于按照一分割规则，对所筛选出来的目标区域进行分割，以将每个目标区域分割成多个子区域；
第二图像获取模块，用于每隔第二预设时间抽取所述监控摄像头所捕获的第二图像；
运动侦测模块，用于将所述第二图像中处于每个子区域中的图像块与前一次抽取的第二图像处于相同子区域的图像块进行比较，以判断各个子区域是否发生了运动事件；及
视频录制模块，用于当某一个子区域发生了运动事件时，控制所述监控摄像头从当前抽取的第二图像开始进行视频录制。
本申请提供的具有计算机可读指令的非易失性存储介质，所述计算机可读指令可被一个或多个处理器执行，以执行如下步骤：
每隔第一预设时间段抽取一监控摄像头所捕获的第一图像；
利用一预先建立的模型，对抽取的第一图像进行区域检测，以提取出包含目标物部分部位或者全部部位的兴趣区域；
利用一分析规则，对所述兴趣区域执行运动区域筛选，以筛选出目标区域；
按照一分割规则，对所筛选出来的目标区域进行分割，以将每个目标区域分割成多个子区域；
每隔第二预设时间抽取所述监控摄像头所捕获的第二图像；
将所述第二图像中处于每个子区域中的图像块与前一次抽取的第二图像处于相同子区域的图像块进行比较，以判断各个子区域是否发生了运动事件；及
当某一个子区域发生了运动事件时，控制所述监控摄像头从当前抽取的第二图像开始进行视频录制。
本申请所述视频录制方法及适用于该方法的服务器、系统及存储介质，可以减少不必要的视频帧的录制，减少存储空间和网络带宽的浪费。
附图说明
图1是本申请视频录制系统第一较佳实施例的服务器应用环境示意图。
图2是本申请视频录制系统第二较佳实施例的终端应用环境示意图。
图3是本申请视频录制系统较佳实施例的功能模块图。
图4是本申请视频录制方法较佳实施例的方法实施流程图。
图5是图4所述视频录制方法较佳实施例中判断各个子区域是否发生了运动事件的详细实施流程图。
具体实施方式
参阅图1所示，是本申请视频录制系统第一较佳实施例的服务器应用环境示意图。
本实施例中，所述视频录制系统10可以安装并运行于一服务器中。如图1所示，所述服务器可以是一台监控服务器1。所述监控服务器1可以通过一通讯模块(未图示)与一监控场所2中所安装的一个或者多个监控摄像头3通讯连接。
所述监控场所2可以是学校、幼儿园、商场、医院、公园、城市 广场、地下人行通道等人员较多的场所，也可以是需求安装监控的特殊区域，如家庭、小超市、ATM(Automatic Teller Machine)机等。本实施例中，所述监控场所2为汽车修理厂，如4S店。
所述监控场所2中安装有一个或者多个监控摄像头3。所述监控摄像头3可以是模拟摄像头。所述模拟摄像头可以将视频采集设备产生的模拟视频信号经过特定的视频捕捉卡转换成数字信号，进而将其传输并储存在于监控服务器1中。本实施例中，所述监控摄像头3是网络摄像头。所述网络摄像头固定好后，用网线与路由器连接，通过路由器与监控服务器1通讯连接，以由监控服务器1进行视频输出。
所述监控服务器1可以包括有处理器以及存储设备(未图示)。所述处理器是运算核心(Core Unit)和控制核心(Control Unit)，用于解释计算机指令以及处理计算机软件中的数据。所述存储设备上存储有数据库、操作系统及上述视频录制系统10。在一些实施例中，存储设备包括有内存储器及非易失性存储介质；视频录制系统、操作系统及数据库存储在非易失性存储介质上；内存储器则为操作系统、数据库和视频录制系统10提供高速缓存的运行环境。
本实施例中，所述视频录制系统10包括至少一个计算机可执行的程序指令代码，该程序指令代码可以被处理器的执行下，实现下述操作。
每隔第一预设时间段，如每隔5分钟，抽取一监控摄像头3所捕获的第一图像；利用一预先建立的模型，对抽取的第一图像进行区域检测，以提取出包含目标物，如车辆部分部位或者全部部位的兴趣区域；利用一分析规则，对所述兴趣区域执行运动区域筛选，以筛选出目标区域；按照一分割规则，对所筛选出来的目标区域进行分割，以将每个目标区域分割成多个子区域；每隔第二预设时间，如0.5秒钟，抽取所述监控摄像头3所捕获的第二图像；将所述第二图像中处于每个子区域中的图像块与前一次抽取的第二图像处于相同子区域的图像块进行比较，以判断各个子区域是否发生了运动事件；当某一个子区域发生了运动事件时，控制所述监控摄像头3从当前抽取的第二图像开始进行视频录制。
其中，所述预先建立的模型为卷积神经网络(Convolutional Neural Network，CNN)模型。
所述模型生成步骤包括：
从各个车辆修理厂数据库中获取各个车辆修理工位监控设备采集的预设数量(例如，10万张)修理工位图像；由于目前修理厂的监控视频已经有很多，本实施例中，可以从中筛选动态视频，并提取关键帧，从而获取大量图像。对采集的各个照片中车辆所处区域进行标注，其中，在标注的过程中，可以用矩形框标注车辆位置，并且给 出标注。这一过程可以采用众包或者公司内部人工标注的形式进行，标注的车辆所处区域即是兴趣区域。利用标注车辆所处区域后的预设数量图像，训练预设模型结构的CNN模型，以生成识别图像中所述兴趣区域的CNN模型。训练的目的是优化CNN网络内各权重的值，使得网络模型作为整体可实际较好地适用于所述兴趣区域的识别。网络模型总共有七层，分别是五个卷积层、一个降采样层和一个全连接层。其中，卷积层由很多个特征向量构造的特征图形成，而特征图的作用就是利用卷积滤波器提取关键特征。降采样层的作用是通过采样方法，去除重复表达的特征点，减少特征提取的数量，从而提高网络层间数据通信效率，可用的采样方法包括最大采样法、均值采样法、随机采样法。全连接层的作用是连接前面的卷积层与降采样，并计算权重矩阵，用于后续的实际分类。图像进入模型后，在每一层均经过前向迭代与后向迭代两个过程，每一次迭代生成一个概率分布，多次迭代后的概率分布进行叠加，系统选取概率分布中取得最大值的类别作为最终的分类结果。
其中，所述分析规则为：分析是否有提取的兴趣区域处于预设的像素区域范围内，例如，预设的像素区域范围包括横坐标区域范围和纵坐标区域范围，其中，所述横坐标区域范围为(X1，X2)，所述纵坐标区域范围为(Y1，Y2)，所述X1代表的是第X1列像素点，所述X2代表的是第X2列像素点，X1小于X2，所述Y1代表的是第Y1行像素点，所述Y2代表的是第Y2行像素点，且Y1小于Y2；若有提取的兴趣区域处于预设的像素区域范围内，则确认该兴趣区域为所述目标区域。该分析规则的原理为：修理厂的监控一般都会对准一个修理工位，保证车辆占据镜头的主要区域“即中间区域”，因此，所述预设的像素区域范围需尽量涵盖镜头的主要区域；该范围不宜过大，防止多个兴趣区域落入其中；该范围亦不宜过小，防止目标区域难以落入其中；可以通过人工的方式校验所述横坐标区域范围和纵坐标区域范围，若过大，则调小，若过小，则调大。
其中，所述分割规则为：采用均匀分割方式，即分割的子区域面积大小、形状均一致，对所述目标区域进行子区域分割；将所述目标区域分割成N*N个子区域，其中，N为大于1的正整数，例如，8*8。采用N*N的子区域进行运动检测而不是针对整体目标进行检测主要有两点考虑：第一是精度，若只是针对目标整体进行像素值比较，可能较小的动作会被其他静态的部位平均掉，导致无法检测到这样的细微动作；第二是效率，可能动作只是发生在某一个区域，则只要检测出这个子区域有动作，即可保存该帧，没有必要继续检测其他部位。例如，在一个例子中，以8*8个子区域为例，若在第一个子区域上检测到了动作，则不需要检测剩下的63个子区域，从而能将效率提高 64倍。
其中，所述判断各个子区域是否发生了运动事件的步骤包括：将当前抽取的第二图像处于每个子区域中的图像块各个像素点的像素值，和前一次抽取的第二图像处于相同子区域的图像块对应像素点的像素值求差值；对每个子区域中的图像块对应的所有差值求和，并将计算的和除以所述图像块的像素点数量，以获得每个子区域中的图像块对应的差值平均值；及若有子区域中的图像块对应的差值平均值大于预设阈值，则确定该子区域发生了运动事件。
在本发明其他较佳实施例中，所述视频录制系统10也可以安装并运行于任何一个终端设备中，如图2所示的移动终端4。所述移动终端4可以是任何具有一定数据处理功能的电子设备，例如智能手机、平板电脑、笔记本电脑、穿戴式手表、穿戴式眼镜等。类似地，终端设备2也包括有处理器以及存储设备(未图示)，所述视频录制系统10包括存储于终端设备2的存储设备的至少一个计算机可执行的程序指令代码；该程序指令代码在所述终端设备2的处理器的执行下，实现实施例一中所述的操作。
本领域技术人员可以理解，图1及图2中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案的服务器或终端设备的限定，具体地，电子设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
可以理解，上述实施例中的非易失性存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)等非易失性存储介质，或随机存储记忆体(Random Access Memory，RAM)等。
可以理解，所述存储设备可以内置或者外接于监控服务器1或终端设备2。
参阅图3所示，是本发明视频录制系统较佳实施例的功能模块图。
所述视频录制系统10的程序代码根据其不同的功能，可以划分为多个功能模块。本发明较佳实施例中，所述视频录制系统10可以包括第一图像获取模块100、建模模块101、筛选模块102、分割模块103、第二图像获取模块104、运动侦测模块105及视频录制模块106。
所述第一图像获取模块100用于每隔第一预设时间段，如每隔5分钟，抽取一监控摄像头3所捕获的第一图像。
所述建模模块101用于利用一预先建立的模型，对抽取的第一图像进行区域检测，以提取出包含目标如，如车辆部分部位或者全部部位的兴趣区域。
其中，所述预先建立的模型为卷积神经网络(Convolutional  Neural Network，CNN)模型。
所述预设类型模型生成步骤包括：
从各个车辆修理厂数据库中获取各个车辆修理工位监控设备采集的预设数量(例如，10万张)修理工位图像；由于目前修理厂的监控视频已经有很多，本实施例中，可以从中筛选动态视频，并提取关键帧，从而获取大量图像。对采集的各个照片中车辆所处区域进行标注，其中，在标注的过程中，可以用矩形框标注车辆位置，并且给出标注。这一过程可以采用众包或者公司内部人工标注的形式进行，标注的车辆所处区域即是兴趣区域。利用标注车辆所处区域后的预设数量图像，训练预设模型结构的CNN模型，以生成识别图像中所述兴趣区域的CNN模型。训练的目的是优化CNN网络内各权重的值，使得网络模型作为整体可实际较好地适用于所述兴趣区域的识别。网络模型总共有七层，分别是五个卷积层、一个降采样层和一个全连接层。其中，卷积层由很多个特征向量构造的特征图形成，而特征图的作用就是利用卷积滤波器提取关键特征。降采样层的作用是通过采样方法，去除重复表达的特征点，减少特征提取的数量，从而提高网络层间数据通信效率，可用的采样方法包括最大采样法、均值采样法、随机采样法。全连接层的作用是连接前面的卷积层与降采样，并计算权重矩阵，用于后续的实际分类。图像进入模型后，在每一层均经过前向迭代与后向迭代两个过程，每一次迭代生成一个概率分布，多次迭代后的概率分布进行叠加，系统选取概率分布中取得最大值的类别作为最终的分类结果。
所述筛选模块102用于利用一分析规则，对所述兴趣区域执行运动区域筛选，以筛选出目标区域。
其中，所述分析规则为：分析是否有提取的兴趣区域处于预设的像素区域范围内，例如，预设的像素区域范围包括横坐标区域范围和纵坐标区域范围，其中，所述横坐标区域范围为(X1，X2)，所述纵坐标区域范围为(Y1，Y2)；若有提取的兴趣区域处于预设的像素区域范围内，则确认该兴趣区域为所述目标区域。该分析规则的原理为：修理厂的监控一般都会对准一个修理工位，保证车辆占据镜头的主要区域“即中间区域”，因此，所述预设的像素区域范围需尽量涵盖镜头的主要区域；该范围不宜过大，防止多个兴趣区域落入其中；该范围亦不宜过小，防止目标区域难以落入其中；可以通过人工的方式校验所述横坐标区域范围和纵坐标区域范围，若过大，则调小，若过小，则调大。
所述分割模块103用于按照一分割规则，对所筛选出来的目标区域进行分割，以将每个目标区域分割成多个子区域。
其中，所述分割规则为：采用均匀分割方式，即分割的子区域面 积大小、形状均一致，对所述目标区域进行子区域分割；将所述目标区域分割成N*N个子区域，其中，N为大于1的正整数，例如，8*8。采用N*N的子区域进行运动检测而不是针对整体目标进行检测主要有两点考虑：第一是精度，若只是针对目标整体进行像素值比较，可能较小的动作会被其他静态的部位平均掉，导致无法检测到这样的细微动作；第二是效率，可能动作只是发生在某一个区域，则只要检测出这个子区域有动作，即可保存该帧，没有必要继续检测其他部位。例如，在一个例子中，以8*8个子区域为例，若在第一个子区域上检测到了动作，则不需要检测剩下的63个子区域，从而能将效率提高64倍。
所述第二图像获取模块104用于每隔第二预设时间，如0.5秒钟，抽取所述监控摄像头3所捕获的第二图像。
所述运动侦测模块105用于将所述第二图像中处于每个子区域中的图像块与前一次抽取的第二图像处于相同子区域的图像块进行比较，以判断各个子区域是否发生了运动事件。
其中，所述判断各个子区域是否发生了运动事件的步骤包括：将当前抽取的第二图像处于其中一个子区域中的图像块的各个像素点的像素值和前一次抽取的第二图像处于相同子区域的图像块对应像素点的像素值求差值；对该子区域中的图像块对应的所有差值求和，并将计算的和除以所述图像块的像素点数量，以获得该子区域中的图像块对应的差值平均值；及若有子区域中的图像块对应的差值平均值大于预设阈值，则确定该子区域发生了运动事件。
所述视频录制模块106用于当某一个子区域发生了运动事件时，控制所述监控摄像头3从当前抽取的第二图像开始进行视频录制。
参阅图4所示，是本发明视频录制方法较佳实施例的方法实施流程图。本实施例所述视频录制方法并不限于流程图中所示步骤，此外流程图中所示步骤中，某些步骤可以省略、步骤之间的顺序可以改变。
步骤S10，第一图像获取模块100每隔第一预设时间段，如每隔5分钟，抽取一监控摄像头3所捕获的第一图像。
步骤S11，建模模块101利用一预先建立的模型，对抽取的第一图像进行区域检测，以提取出包含目标物，如车辆部分部位或者全部部位的兴趣区域。
其中，所述预先建立的模型为卷积神经网络(Convolutional Neural Network，CNN)模型。
所述预设类型模型生成步骤包括：
从各个车辆修理厂数据库中获取各个车辆修理工位监控设备采集的预设数量(例如，10万张)修理工位图像；由于目前修理厂的监控视频已经有很多，本实施例中，可以从中筛选动态视频，并提取 关键帧，从而获取大量图像。对采集的各个照片中车辆所处区域进行标注，其中，在标注的过程中，可以用矩形框标注车辆位置，并且给出标注。这一过程可以采用众包或者公司内部人工标注的形式进行，标注的车辆所处区域即是兴趣区域。利用标注车辆所处区域后的预设数量图像，训练预设模型结构的CNN模型，以生成识别图像中所述兴趣区域的CNN模型。训练的目的是优化CNN网络内各权重的值，使得网络模型作为整体可实际较好地适用于所述兴趣区域的识别。网络模型总共有七层，分别是五个卷积层、一个降采样层和一个全连接层。其中，卷积层由很多个特征向量构造的特征图形成，而特征图的作用就是利用卷积滤波器提取关键特征。降采样层的作用是通过采样方法，去除重复表达的特征点，减少特征提取的数量，从而提高网络层间数据通信效率，可用的采样方法包括最大采样法、均值采样法、随机采样法。全连接层的作用是连接前面的卷积层与降采样，并计算权重矩阵，用于后续的实际分类。图像进入模型后，在每一层均经过前向迭代与后向迭代两个过程，每一次迭代生成一个概率分布，多次迭代后的概率分布进行叠加，系统选取概率分布中取得最大值的类别作为最终的分类结果。
步骤S12，筛选模块102利用一分析规则，对所述兴趣区域执行运动区域筛选，以筛选出目标区域。
其中，所述分析规则为：分析是否有提取的兴趣区域处于预设的像素区域范围内，例如，预设的像素区域范围包括横坐标区域范围和纵坐标区域范围，其中，所述横坐标区域范围为(X1，X2)，所述纵坐标区域范围为(Y1，Y2)；若有提取的兴趣区域处于预设的像素区域范围内，则确认该兴趣区域为所述目标区域。该分析规则的原理为：修理厂的监控一般都会对准一个修理工位，保证车辆占据镜头的主要区域“即中间区域”，因此，所述预设的像素区域范围需尽量涵盖镜头的主要区域；该范围不宜过大，防止多个兴趣区域落入其中；该范围亦不宜过小，防止目标区域难以落入其中；可以通过人工的方式校验所述横坐标区域范围和纵坐标区域范围，若过大，则调小，若过小，则调大。
步骤S13，筛选模块102判断是否筛选出至少一个目标区域。当没有筛选出任何目标区域，则返回上述的步骤10，重新执行第一图像的抽取。当筛选出了目标区域时，执行下述的步骤S14。
步骤S14，分割模块103按照一分割规则，对所筛选出来的目标区域进行分割，以将每个目标区域分割成多个子区域。
其中，所述分割规则为：采用均匀分割方式，即分割的子区域面积大小、形状均一致，对所述目标区域进行子区域分割；将所述目标区域分割成N*N个子区域，其中，N为大于1的正整数，例如，8*8。 采用N*N的子区域进行运动检测而不是针对整体目标进行检测主要有两点考虑：第一是精度，若只是针对目标整体进行像素值比较，可能较小的动作会被其他静态的部位平均掉，导致无法检测到这样的细微动作；第二是效率，可能动作只是发生在某一个区域，则只要检测出这个子区域有动作，即可保存该帧，没有必要继续检测其他部位。例如，在一个例子中，以8*8个子区域为例，若在第一个子区域上检测到了动作，则不需要检测剩下的63个子区域，从而能将效率提高64倍。
步骤S15，第二图像获取模块104每隔第二预设时间，如0.5秒钟，抽取所述监控摄像头3所捕获的第二图像。
步骤S16，运动侦测模块105将所述第二图像中处于每个子区域中的图像块与前一次抽取的第二图像处于相同子区域的图像块进行比较，以判断各个子区域是否发生了运动事件。步骤S16的详细实施流程参见下述的图5中的描述。
步骤S17，运动侦测模块105判断各个子区域是否发生了运动事件。当没有任何子区域发生了运动事件，则返回上述的步骤S15。当有任何子区域发生了运动事件，则执行下述的步骤S18。
步骤S18，视频录制模块106控制所述监控摄像头3从当前抽取的第二图像开始进行视频录制。
参阅图5所示，是图4中步骤S16，即判断各个子区域是否发生了运动事件的详细实施流程图。本实施例所述视频录制方法并不限于流程图中所示步骤，此外流程图中所示步骤中，某些步骤可以省略、步骤之间的顺序可以改变。
步骤S160，运动侦测模块105获取当前抽取的第二图像处于其中一个子区域中的图像块的各个像素点的像素值。
步骤S161，运动侦测模块105将当前抽取的第二图像处于所述子区域中的图像块各个像素点的像素值和前一次抽取的第二图像处于相同子区域的图像块对应像素点的像素值求差值。
步骤S162，运动侦测模块105对所述子区域中的图像块对应的所有差值求和，并将计算的和除以所述图像块的像素点数量，以获得所述子区域中的图像块对应的差值平均值。
步骤S163，运动侦测模块105判断该子区域的图像块对应的差值平均值是否大于预设阈值。当该子区域的图像块对应的差值平均值大于预设阈值，则执行步骤S164，否则，当该子区域的图像块对应的差值平均值小于预设阈值，则执行步骤S165。
步骤S164，运动侦测模块105确定所述子区域发生了运动事件。
步骤S165，运动侦测模块105确定所述子区域没有发生运动事件，并返回上述的步骤S160，运动侦测模块105获取当前抽取的第 二图像处于下一个子区域中的图像块的各个像素点的像素值。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的程序可存储于一计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，所述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)等非易失性存储介质，或随机存储记忆体(Random Access Memory，RAM)等。
以上所述实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
最后所应说明的是，以上实施例仅用以说明本发明的技术方案而非限制，尽管参照较佳实施例对本发明进行了详细说明，本领域的普通技术人员应当理解，可以对本发明的技术方案进行修改或等同替换，而不脱离本发明技术方案的精神和范围。

Claims (20)

  1. 一种视频录制方法，其特征在于，该方法包括：
    每隔第一预设时间段抽取一监控摄像头所捕获的第一图像；
    利用一预先建立的模型，对抽取的第一图像进行区域检测，以提取出包含目标物部分部位或者全部部位的兴趣区域；
    利用一分析规则，对所述兴趣区域执行运动区域筛选，以筛选出目标区域；
    按照一分割规则，对所筛选出来的目标区域进行分割，以将每个目标区域分割成多个子区域；
    每隔第二预设时间抽取所述监控摄像头所捕获的第二图像；
    将所述第二图像中处于每个子区域中的图像块与前一次抽取的第二图像处于相同子区域的图像块进行比较，以判断各个子区域是否发生了运动事件；及
    当某一个子区域发生了运动事件时，控制所述监控摄像头从当前抽取的第二图像开始进行视频录制。
  2. 如权利要求1所述的方法，其特征在于，所述预先建立的模型为卷积神经网络模型。
  3. 如权利要求1所述的方法，其特征在于，所述分析规则为：
    分析是否有提取的兴趣区域处于预设的像素区域范围内，所述预设的像素区域范围包括横坐标区域范围和纵坐标区域范围，其中，所述横坐标区域范围为(X1，X2)，所述纵坐标区域范围为(Y1，Y2)，所述X1代表的是第X1列像素点，所述X2代表的是第X2列像素点，X1小于X2，所述Y1代表的是第Y1行像素点，所述Y2代表的是第Y2行像素点，且Y1小于Y2；
    若有提取的兴趣区域处于预设的像素区域范围内，则确认该兴趣区域为所述目标区域。
  4. 如权利要求1所述的方法，其特征在于，所述分割规则为：采用均匀分割方式对所述目标区域进行子区域分割；将所述目标区域分割成N*N个子区域，其中，N为大于1的正整数。
  5. 如权利要求1所述的方法，其特征在于，所述判断各个子区域是否发生了运动事件的步骤包括：
    将当前抽取的第二图像处于每个子区域中的图像块中的各个像素点的像素值和前一次抽取的第二图像处于相同子区域的图像块对应像素点的像素值求差值；
    对每个子区域中的图像块对应的所有差值求和，并将计算的和除以所述图像块的像素点数量，以获得每个子区域中的图像块对应的差值平均值；及
    若有子区域中的图像块对应的差值平均值大于预设阈值，则确定该子区域发生了运动事件。
  6. 一种服务器，其特征在于，该服务器包括存储设备以及处理器，其中：
    所述存储设备，用于存储一个视频录制系统；
    所述处理器，用于执行所述视频录制系统，以执行如下步骤：
    每隔第一预设时间段抽取一监控摄像头所捕获的第一图像；
    利用一预先建立的模型，对抽取的第一图像进行区域检测，以提取出包含目标物部分部位或者全部部位的兴趣区域；
    利用一分析规则，对所述兴趣区域执行运动区域筛选，以筛选出目标区域；
    按照一分割规则，对所筛选出来的目标区域进行分割，以将每个目标区域分割成多个子区域；
    每隔第二预设时间抽取所述监控摄像头所捕获的第二图像；
    将所述第二图像中处于每个子区域中的图像块与前一次抽取的第二图像处于相同子区域的图像块进行比较，以判断各个子区域是否发生了运动事件；及
    当某一个子区域发生了运动事件时，控制所述监控摄像头从当前抽取的第二图像开始进行视频录制。
  7. 如权利要求6所述的服务器，其特征在于，所述预先建立的模型为卷积神经网络模型。
  8. 如权利要求6所述的服务器，其特征在于，所述分析规则为：分析是否有提取的兴趣区域处于预设的像素区域范围内，所述预设的像素区域范围包括横坐标区域范围和纵坐标区域范围，其中，所述横坐标区域范围为(X1，X2)，所述纵坐标区域范围为(Y1，Y2)；若有提取的兴趣区域处于预设的像素区域范围内，则确认该兴趣区域为所述目标区域。
  9. 如权利要求6所述的服务器，其特征在于，所述分割规则为：采用均匀分割方式对所述目标区域进行子区域分割；将所述目标区域分割成N*N个子区域，其中，N为大于1的正整数。
  10. 如权利要求6所述的服务器，其特征在于，所述判断各个子区 域是否发生了运动事件的步骤包括：
    将当前抽取的第二图像处于每个子区域中的图像块各个像素点的像素值，和前一次抽取的第二图像处于相同子区域的图像块对应像素点的像素值求差值；
    对每个子区域中的图像块对应的所有差值求和，并将计算的和除以所述图像块的像素点数量，以获得每个子区域中的图像块对应的差值平均值；及
    若有子区域中的图像块对应的差值平均值大于预设阈值，则确定该子区域发生了运动事件。
  11. 一种视频录制系统，其特征在于，该系统包括：
    第一图像获取模块，用于每隔第一预设时间段抽取一监控摄像头所捕获的第一图像；
    建模模块，用于利用一预先建立的模型，对抽取的第一图像进行区域检测，以提取出包含目标物部分部位或者全部部位的兴趣区域；
    筛选模块，用于利用一分析规则，对所述兴趣区域执行运动区域筛选，以筛选出目标区域；
    分割模块，用于按照一分割规则，对所筛选出来的目标区域进行分割，以将每个目标区域分割成多个子区域；
    第二图像获取模块，用于每隔第二预设时间抽取所述监控摄像头所捕获的第二图像；
    运动侦测模块，用于将所述第二图像中处于每个子区域中的图像块与前一次抽取的第二图像处于相同子区域的图像块进行比较，以判断各个子区域是否发生了运动事件；及
    视频录制模块，用于当某一个子区域发生了运动事件时，控制所述监控摄像头从当前抽取的第二图像开始进行视频录制。
  12. 如权利要求11所述的系统，其特征在于，所述预先建立的模型为卷积神经网络模型。
  13. 如权利要求11所述的系统，其特征在于，所述分析规则为：
    分析是否有提取的兴趣区域处于预设的像素区域范围内，所述预设的像素区域范围包括横坐标区域范围和纵坐标区域范围，其中，所述横坐标区域范围为(X1，X2)，所述纵坐标区域范围为(Y1，Y2)，所述X1代表的是第X1列像素点，所述X2代表的是第X2列像素点，X1小于X2，所述Y1代表的是第Y1行像素点，所述Y2代表的是第Y2行像素点，且Y1小于Y2；
    若有提取的兴趣区域处于预设的像素区域范围内，则确认该兴趣 区域为所述目标区域。
  14. 如权利要求11所述的系统，其特征在于，所述分割规则为：采用均匀分割方式对所述目标区域进行子区域分割；将所述目标区域分割成N*N个子区域，其中，N为大于1的正整数。
  15. 如权利要求11所述的系统，其特征在于，所述运动侦测模块用于：
    将当前抽取的第二图像处于每个子区域中的图像块中的各个像素点的像素值和前一次抽取的第二图像处于相同子区域的图像块对应像素点的像素值求差值；
    对每个子区域中的图像块对应的所有差值求和，并将计算的和除以所述图像块的像素点数量，以获得每个子区域中的图像块对应的差值平均值；及
    若有子区域中的图像块对应的差值平均值大于预设阈值，则确定该子区域发生了运动事件。
  16. 一种具有计算机可读指令的存储介质，所述计算机可读指令可被一个或多个处理器执行，以执行如下步骤：
    每隔第一预设时间段抽取一监控摄像头所捕获的第一图像；
    利用一预先建立的模型，对抽取的第一图像进行区域检测，以提取出包含目标物部分部位或者全部部位的兴趣区域；
    利用一分析规则，对所述兴趣区域执行运动区域筛选，以筛选出目标区域；
    按照一分割规则，对所筛选出来的目标区域进行分割，以将每个目标区域分割成多个子区域；
    每隔第二预设时间抽取所述监控摄像头所捕获的第二图像；
    将所述第二图像中处于每个子区域中的图像块与前一次抽取的第二图像处于相同子区域的图像块进行比较，以判断各个子区域是否发生了运动事件；及
    当某一个子区域发生了运动事件时，控制所述监控摄像头从当前抽取的第二图像开始进行视频录制。
  17. 如权利要求16所述的存储介质，其特征在于，所述预先建立的模型为卷积神经网络模型。
  18. 如权利要求16所述的存储介质，其特征在于，所述分析规则为：
    分析是否有提取的兴趣区域处于预设的像素区域范围内，所述预设的像素区域范围包括横坐标区域范围和纵坐标区域范围，其中，所 述横坐标区域范围为(X1，X2)，所述纵坐标区域范围为(Y1，Y2)，所述X1代表的是第X1列像素点，所述X2代表的是第X2列像素点，X1小于X2，所述Y1代表的是第Y1行像素点，所述Y2代表的是第Y2行像素点，且Y1小于Y2；
    若有提取的兴趣区域处于预设的像素区域范围内，则确认该兴趣区域为所述目标区域。
  19. 如权利要求16所述的存储介质，其特征在于，所述分割规则为：采用均匀分割方式对所述目标区域进行子区域分割；将所述目标区域分割成N*N个子区域，其中，N为大于1的正整数。
  20. 如权利要求16所述的存储介质，其特征在于，所述判断各个子区域是否发生了运动事件的步骤包括：
    将当前抽取的第二图像处于每个子区域中的图像块中的各个像素点的像素值和前一次抽取的第二图像处于相同子区域的图像块对应像素点的像素值求差值；
    对每个子区域中的图像块对应的所有差值求和，并将计算的和除以所述图像块的像素点数量，以获得每个子区域中的图像块对应的差值平均值；及
    若有子区域中的图像块对应的差值平均值大于预设阈值，则确定该子区域发生了运动事件。
PCT/CN2017/080113 2016-04-14 2017-04-11 视频录制方法、服务器、系统及存储介质 WO2017177902A1 (zh)

Priority Applications (6)

Application Number Priority Date Filing Date Title
SG11201800364YA SG11201800364YA (en) 2016-04-14 2017-04-11 Video recording system, server, system, and storage medium
US15/737,323 US10349003B2 (en) 2016-04-14 2017-04-11 Video recording system, server, system, and storage medium
KR1020187019521A KR102155182B1 (ko) 2016-04-14 2017-04-11 비디오 리코딩 방법, 서버, 시스템 및 저장 매체
EP17781878.8A EP3445044B1 (en) 2016-04-14 2017-04-11 Video recording method, server, system, and storage medium
AU2017250159A AU2017250159B2 (en) 2016-04-14 2017-04-11 Video recording method, server, system, and storage medium
JP2018524835A JP6425856B1 (ja) 2016-04-14 2017-04-11 ビデオ録画方法、サーバー、システム及び記憶媒体

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610234956.7 2016-04-14
CN201610234956.7A CN106027931B (zh) 2016-04-14 2016-04-14 视频录制方法及服务器

Publications (1)

Publication Number Publication Date
WO2017177902A1 true WO2017177902A1 (zh) 2017-10-19

Family

ID=57081964

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/080113 WO2017177902A1 (zh) 2016-04-14 2017-04-11 视频录制方法、服务器、系统及存储介质

Country Status (8)

Country Link
US (1) US10349003B2 (zh)
EP (1) EP3445044B1 (zh)
JP (1) JP6425856B1 (zh)
KR (1) KR102155182B1 (zh)
CN (1) CN106027931B (zh)
AU (1) AU2017250159B2 (zh)
SG (1) SG11201800364YA (zh)
WO (1) WO2017177902A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110502986A (zh) * 2019-07-12 2019-11-26 平安科技（深圳）有限公司 识别图像中人物位置方法、装置、计算机设备和存储介质
CN111339879A (zh) * 2020-02-19 2020-06-26 安徽领云物联科技有限公司 一种兵器室单人入室监测方法及装置
CN111507896A (zh) * 2020-04-27 2020-08-07 北京字节跳动网络技术有限公司 图像液化处理方法、装置、设备和存储介质
US10949952B2 (en) * 2018-06-07 2021-03-16 Beijing Kuangshi Technology Co., Ltd. Performing detail enhancement on a target in a denoised image

Families Citing this family (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106027931B (zh) 2016-04-14 2018-03-16 平安科技（深圳）有限公司 视频录制方法及服务器
CN107766829A (zh) * 2017-10-27 2018-03-06 浙江大华技术股份有限公司 一种物品检测的方法和设备
GB2569555B (en) * 2017-12-19 2022-01-12 Canon Kk Method and apparatus for detecting deviation from a motion pattern in a video
GB2569557B (en) 2017-12-19 2022-01-12 Canon Kk Method and apparatus for detecting motion deviation in a video
GB2569556B (en) * 2017-12-19 2022-01-12 Canon Kk Method and apparatus for detecting motion deviation in a video sequence
CN109522814B (zh) * 2018-10-25 2020-10-02 清华大学 一种基于视频数据的目标追踪方法及装置
CN112019868A (zh) * 2019-05-31 2020-12-01 广州虎牙信息科技有限公司 人像分割方法、装置及电子设备
JP7294927B2 (ja) * 2019-07-23 2023-06-20 ファナック株式会社 相違点抽出装置
KR102090739B1 (ko) * 2019-10-21 2020-03-18 주식회사 휴머놀러지 영상의 유사도 분석을 위해 영상 영역 격자 다분할을 이용하는 지능형 이동 감시 시스템 및 그 감시방법
CN111652128B (zh) * 2020-06-02 2023-09-01 浙江大华技术股份有限公司 一种高空电力作业安全监测方法、系统和存储装置
CN112203054B (zh) * 2020-10-09 2022-10-14 深圳赛安特技术服务有限公司 监控视频点位标注方法、装置、存储介质及电子设备
CN112601049B (zh) * 2020-12-08 2023-07-25 北京精英路通科技有限公司 视频的监控方法、装置、计算机设备及存储介质
CN114295058B (zh) * 2021-11-29 2023-01-17 清华大学 一种建筑结构的整面动位移的测量方法
CN114666497B (zh) * 2022-02-28 2024-03-15 青岛海信移动通信技术有限公司 成像方法、终端设备及存储介质
CN115314717B (zh) * 2022-10-12 2022-12-20 深流微智能科技(深圳)有限公司 视频传输方法、装置、电子设备和计算机可读存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101017573A (zh) * 2007-02-09 2007-08-15 南京大学 一种基于视频监控的运动目标检测与识别方法
US20130201338A1 (en) * 2012-02-07 2013-08-08 Sensormatic Electronics, LLC Method and System for Monitoring Portal to Detect Entry and Exit
CN104270619A (zh) * 2014-10-22 2015-01-07 中国建设银行股份有限公司 一种安全告警方法及装置
CN104601918A (zh) * 2014-12-29 2015-05-06 小米科技有限责任公司 视频录制方法和装置
CN105279898A (zh) * 2015-10-28 2016-01-27 小米科技有限责任公司 报警方法及装置
CN106027931A (zh) * 2016-04-14 2016-10-12 平安科技（深圳）有限公司 视频录制方法及服务器

Family Cites Families (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8711217B2 (en) * 2000-10-24 2014-04-29 Objectvideo, Inc. Video surveillance system employing video primitives
CN100426837C (zh) * 2005-05-10 2008-10-15 北京中星微电子有限公司 一种监控摄像方法及装置
JP5134556B2 (ja) * 2009-01-08 2013-01-30 株式会社日立製作所 監視記録装置、監視システムおよび監視記録方法
JP2011035663A (ja) * 2009-07-31 2011-02-17 Panasonic Corp 監視装置および監視方法
JP5358851B2 (ja) * 2009-11-12 2013-12-04 将文 萩原 不審行動検知方法および不審行動検知装置
US8660368B2 (en) * 2011-03-16 2014-02-25 International Business Machines Corporation Anomalous pattern discovery
AU2012227263A1 (en) * 2012-09-21 2014-04-10 Canon Kabushiki Kaisha Differentiating abandoned and removed object using temporal edge information
JP5954106B2 (ja) * 2012-10-22 2016-07-20 ソニー株式会社 情報処理装置、情報処理方法、プログラム、及び情報処理システム
KR102003671B1 (ko) * 2012-10-29 2019-07-25 삼성전자주식회사 영상 처리 방법 및 장치
US9521377B2 (en) * 2013-10-08 2016-12-13 Sercomm Corporation Motion detection method and device using the same
KR102015954B1 (ko) * 2014-03-21 2019-08-29 한화테크윈 주식회사 영상 처리 시스템 및 영상 처리 방법
KR101681233B1 (ko) * 2014-05-28 2016-12-12 한국과학기술원 저 에너지/해상도 가지는 얼굴 검출 방법 및 장치
US20160042621A1 (en) * 2014-06-13 2016-02-11 William Daylesford Hogg Video Motion Detection Method and Alert Management
US10481696B2 (en) * 2015-03-03 2019-11-19 Nvidia Corporation Radar based user interface
US20170076195A1 (en) * 2015-09-10 2017-03-16 Intel Corporation Distributed neural networks for scalable real-time analytics
US10437878B2 (en) * 2016-12-28 2019-10-08 Shutterstock, Inc. Identification of a salient portion of an image

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101017573A (zh) * 2007-02-09 2007-08-15 南京大学 一种基于视频监控的运动目标检测与识别方法
US20130201338A1 (en) * 2012-02-07 2013-08-08 Sensormatic Electronics, LLC Method and System for Monitoring Portal to Detect Entry and Exit
CN104270619A (zh) * 2014-10-22 2015-01-07 中国建设银行股份有限公司 一种安全告警方法及装置
CN104601918A (zh) * 2014-12-29 2015-05-06 小米科技有限责任公司 视频录制方法和装置
CN105279898A (zh) * 2015-10-28 2016-01-27 小米科技有限责任公司 报警方法及装置
CN106027931A (zh) * 2016-04-14 2016-10-12 平安科技（深圳）有限公司 视频录制方法及服务器

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10949952B2 (en) * 2018-06-07 2021-03-16 Beijing Kuangshi Technology Co., Ltd. Performing detail enhancement on a target in a denoised image
CN110502986A (zh) * 2019-07-12 2019-11-26 平安科技（深圳）有限公司 识别图像中人物位置方法、装置、计算机设备和存储介质
CN111339879A (zh) * 2020-02-19 2020-06-26 安徽领云物联科技有限公司 一种兵器室单人入室监测方法及装置
CN111339879B (zh) * 2020-02-19 2023-06-02 安徽领云物联科技有限公司 一种兵器室单人入室监测方法及装置
CN111507896A (zh) * 2020-04-27 2020-08-07 北京字节跳动网络技术有限公司 图像液化处理方法、装置、设备和存储介质
CN111507896B (zh) * 2020-04-27 2023-09-05 抖音视界有限公司 图像液化处理方法、装置、设备和存储介质

Also Published As

Publication number Publication date
US10349003B2 (en) 2019-07-09
JP6425856B1 (ja) 2018-11-21
CN106027931B (zh) 2018-03-16
EP3445044B1 (en) 2020-07-29
KR102155182B1 (ko) 2020-09-11
EP3445044A1 (en) 2019-02-20
US20180227538A1 (en) 2018-08-09
AU2017250159A1 (en) 2017-11-23
JP2018535496A (ja) 2018-11-29
EP3445044A4 (en) 2019-09-18
CN106027931A (zh) 2016-10-12
SG11201800364YA (en) 2018-02-27
KR20180133379A (ko) 2018-12-14
AU2017250159B2 (en) 2018-07-26

Similar Documents

Publication Publication Date Title
WO2017177902A1 (zh) 视频录制方法、服务器、系统及存储介质
CN109815843B (zh) 图像处理方法及相关产品
US9373034B2 (en) Apparatus and method for tracking object
US11790553B2 (en) Method and apparatus for detecting target object, electronic device and storage medium
US20180114071A1 (en) Method for analysing media content
Viswanath et al. Background modelling from a moving camera
WO2020008667A1 (en) System and method for video anomaly detection
CN111079621B (zh) 检测对象的方法、装置、电子设备和存储介质
US20220301317A1 (en) Method and device for constructing object motion trajectory, and computer storage medium
Saran et al. Traffic video surveillance: Vehicle detection and classification
CN113496208B (zh) 视频的场景分类方法及装置、存储介质、终端
CN115760912A (zh) 运动目标跟踪方法、装置、设备及计算机可读存储介质
Angelo A novel approach on object detection and tracking using adaptive background subtraction method
CN113112479A (zh) 基于关键区块提取的渐进式目标检测方法和装置
Jaiswal et al. Survey paper on various techniques of recognition and tracking
CN113869163B (zh) 目标跟踪方法、装置、电子设备及存储介质
CN116349233A (zh) 用于确定视频中的运动显著性和视频回放风格的方法和电子装置
SanMiguel et al. Object-size invariant anomaly detection in video-surveillance
EP2570962A1 (en) Video analysis
Puri et al. Ssd based vehicle number plate detection and recognition
Dinakaran et al. Image resolution impact analysis on pedestrian detection in smart cities surveillance
CN114820692B (zh) 跟踪目标的状态分析方法、装置、存储介质和终端
Chandrajit et al. Data Association and Prediction for Tracking Multiple Objects
CN111046724B (zh) 一种基于区域匹配网络的行人检索方法
Asundi et al. Raindrop detection algorithm for ADAS

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2017250159

Country of ref document: AU

Date of ref document: 20170411

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: 15737323

Country of ref document: US

WWE Wipo information: entry into national phase

Ref document number: 11201800364Y

Country of ref document: SG

WWE Wipo information: entry into national phase

Ref document number: 2018524835

Country of ref document: JP

ENP Entry into the national phase

Ref document number: 20187019521

Country of ref document: KR

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17781878

Country of ref document: EP

Kind code of ref document: A1