WO2022198897A1 - 一种路侧停车的管理方法及装置 - Google Patents

一种路侧停车的管理方法及装置 Download PDF

Info

Publication number
WO2022198897A1
WO2022198897A1 PCT/CN2021/112931 CN2021112931W WO2022198897A1 WO 2022198897 A1 WO2022198897 A1 WO 2022198897A1 CN 2021112931 W CN2021112931 W CN 2021112931W WO 2022198897 A1 WO2022198897 A1 WO 2022198897A1
Authority
WO
WIPO (PCT)
Prior art keywords
motion vector
image
vehicle
interest
region
Prior art date
Application number
PCT/CN2021/112931
Other languages
English (en)
French (fr)
Inventor
闫军
王永飞
Original Assignee
超级视线科技有限公司
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 超级视线科技有限公司 filed Critical 超级视线科技有限公司
Publication of WO2022198897A1 publication Critical patent/WO2022198897A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/14—Traffic control systems for road vehicles indicating individual free spaces in parking areas
    • G08G1/145—Traffic control systems for road vehicles indicating individual free spaces in parking areas where the indication depends on the parking areas
    • G08G1/148—Management of a network of parking areas
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/017—Detecting movement of traffic to be counted or controlled identifying vehicles
    • G08G1/0175—Detecting movement of traffic to be counted or controlled identifying vehicles by photographing vehicles, e.g. when violating traffic rules

Definitions

  • the invention relates to the technical field of intelligent transportation, and in particular, to a management method and device for on-street parking.
  • the roadside parking management technology that has been commercialized in some cities mainly includes geomagnetic, low-level road video piles, and high-level video.
  • High-level video has rich forensic information, complete evidence chain, high installation location, and is not easy to be damaged. Scene adaptability Strong and reliable, it has become a priority on-street parking management method in major cities.
  • vehicle target detection and tracking are continuously performed on each frame of real-time image collected by the video camera, and license plate recognition is performed on each frame to obtain parking event-related information and vehicle-related information.
  • the parking management device needs to continuously track and process each frame of the video camera in real time.
  • the parking behavior itself is an intermittent behavior, and there must be a certain period of time.
  • this method can only continuously analyze the video image, but the process of video image analysis requires the use of deep learning target detection and tracking algorithms, which requires edge AI consumption
  • the computing host has a lot of computing power, so it will not be able to process more video camera images in parallel, making this method unable to manage more berths at the same time.
  • This method performs differential comparison on each frame of image collected by the video camera. When it is found that the parking target area has a change exceeding the image differential threshold, a more in-depth video image analysis and processing is performed on the parking process. Since this method uses the differential detection algorithm, the computing power consumption of the differential algorithm is small, which avoids the uninterrupted vehicle target detection and real-time tracking computing power for the real-time image of each frame of the video camera under the technology-video stream real-time detection and tracking mode. To a certain extent, it increases the number of video cameras that can be processed in parallel by a single edge AI computing host. However, the difference algorithm can only detect the size change of the image pixel value, and cannot determine the moving direction of the object. Interfering factors such as pedestrians and passing vehicles will lead to a large number of differential misjudgments, as well as large errors in estimating the time interval of the parking event. This has a serious impact on subsequent video image analysis.
  • a distance sensor such as radar sensor, ultrasonic sensor, etc.
  • the monitoring device When a vehicle appears in the parking target area and the distance sensor detects a change in the distance of the target area, the monitoring device will be notified to take a snapshot, and Recognize vehicle license plate information for vehicle management of parking spaces.
  • this management method needs to add distance sensors, which makes the management cost relatively high, and in the environment such as rain and snow, the distance accuracy error measured by the distance sensor is extremely large, and the anti-interference ability is poor, which greatly reduces the detection accuracy.
  • the embodiments of the present invention provide a management method and device for on-street parking. Under the premise of ensuring accurate analysis of parking events, the utilization rate of the computing power of the host is greatly improved, thereby realizing efficient and accurate processing of multiple pole positions and multiple roads. Video camera parking events are analyzed and processed.
  • an embodiment of the present invention provides a method for managing on-street parking, including:
  • the dense optical flow calculation is performed on the region of interest to obtain the motion vector direction and motion vector distance of the region of interest in each frame of image;
  • the average value of the motion vector direction and the average motion vector distance of the region of interest of each frame image in the statistical interval of a predetermined period of time is calculated by dense optical flow ;
  • the average value of the direction of the motion vector and the average value of the distance of the motion vector determine the vehicle suspected of having an exit/entry event in the parking area
  • vehicle information of a vehicle suspected of having an exit/entry event is identified, and whether the vehicle has an exit/entry event is determined according to the vehicle information.
  • the steps include:
  • the berth area in the video frame image captured by the video of each camera is determined.
  • the coordinate information of the berth area in each video frame image is determined, and the long side of the rectangular frame of the berth area is extended for a predetermined length along the direction in which the Y-axis value gradually decreases, so as to obtain the coordinate information of each video frame.
  • Region of interest for dense optical flow calculations is determined, and the long side of the rectangular frame of the berth area is extended for a predetermined length along the direction in which the Y-axis value gradually decreases.
  • the dense optical flow calculation is performed on the region of interest to obtain the motion vector direction and motion vector distance of the region of interest in each frame of image, including:
  • Each frame of image is decoded, and dense optical flow calculation is performed on the region of interest in each frame of image after decoding, and the motion vector direction and motion vector distance of the region of interest in each frame of image are obtained.
  • the dense optical flow calculation is performed on the region of interest in each frame of image after decoding, and the motion vector direction and motion vector distance of the region of interest in each frame of image are obtained, including:
  • the region of interest of the decoded current frame is equally divided into a predetermined number of calculation regions
  • the motion vector directions and motion vector distances of the respective calculation regions are averaged to obtain the motion vector directions and motion vector distances of the region of interest in the current frame.
  • the average value of the direction of the motion vector and the average value of the distance of the motion vector to determine the vehicle suspected of having an exit/entry event in the parking area, including:
  • identifying the vehicle information of the vehicle suspected of having an exit/entry event based on the video image information, and determining whether the vehicle has an exit/entry event according to the vehicle information including:
  • the vehicle in the to-be-identified video image information is detected and tracked to determine whether the vehicle has an exit/entry event.
  • an embodiment of the present invention provides an apparatus for managing on-street parking, including:
  • the acquisition and determination module is used to acquire the video image information collected by multiple cameras in real time, and determine the berth area in each video frame image;
  • the expansion module is used to expand the berth area in each video frame image to obtain the area of interest for the dense optical flow calculation of each video frame;
  • the first calculation module is used to perform dense optical flow calculation on the region of interest, and obtain the motion vector direction and motion vector distance of the region of interest of each frame of image;
  • the second calculation module is used to calculate the average motion vector direction of the region of interest of each frame image within a statistical interval of a predetermined duration by dense optical flow according to the obtained motion vector direction and motion vector distance of the region of interest of each frame image. value and motion vector distance average;
  • a first determination module configured to determine vehicles suspected of having an exit/entry event in the parking area according to the average value of the direction of the motion vector and the average value of the distance of the motion vector;
  • the identification and determination module is configured to identify the vehicle information of the vehicle suspected of having an exit/entry event based on the video image information, and determine whether the vehicle has an exit/entry event according to the vehicle information.
  • the second determination module is used to determine the coordinates of each vertex of the vehicle parking space in each image in the predetermined image acquisition area;
  • the third determining module is configured to determine, according to the coordinates, the berth area in the video frame image captured by the video of each camera.
  • expansion module is specifically used for
  • the coordinate information of the berth area in each video frame image is determined, and the long side of the rectangular frame of the berth area is extended for a predetermined length along the direction in which the Y-axis value gradually decreases, so as to obtain the coordinate information of each video frame.
  • Region of interest for dense optical flow calculations is determined, and the long side of the rectangular frame of the berth area is extended for a predetermined length along the direction in which the Y-axis value gradually decreases.
  • the first computing module includes:
  • the decoding unit is used for decoding each frame of image, and performs dense optical flow calculation on the region of interest in each frame of image after decoding, and obtains the motion vector direction and motion vector distance of the region of interest in each frame of image.
  • decoding unit is specifically used for
  • the region of interest of the decoded current frame is equally divided into a predetermined number of calculation regions
  • the motion vector directions and motion vector distances of the respective calculation regions are averaged to obtain the motion vector directions and motion vector distances of the region of interest in the current frame.
  • the first determining module is specifically used for
  • identification and determination module is specifically used for
  • the vehicle in the to-be-identified video image information is detected and tracked to determine whether the vehicle has an exit/entry event.
  • image analysis and calculation can be performed on video information collected by multi-channel video equipment at the same time, and by performing dense optical flow calculation on the region of interest in the dense optical flow calculation in the video information, accurate
  • the motion vector direction and motion vector distance of each frame of image are determined accurately, which provides an important premise guarantee for the subsequent determination of vehicles suspected to have exit/entry events in the berth area.
  • a large number of unnecessary calculation video frame images are filtered, which greatly reduces the calculation amount of subsequent video image analysis to determine whether the vehicle has an exit/entry event.
  • the algorithm requires high real-time performance, thus realizing the efficient and accurate analysis and processing of parking events of multi-pole and multi-channel video cameras; realizing the premise of ensuring the accurate analysis of parking events, greatly improving the computing power of the host computer. Utilization, further, greatly reduces the cost of parking management.
  • FIG. 1 is a flowchart of a management method for on-street parking in an embodiment of the present invention
  • FIG. 2 is a schematic structural diagram of a management device for on-street parking in an embodiment of the present invention.
  • Embodiments of the present invention have the following beneficial effects: through the present invention, image analysis and calculation can be performed on video information collected by multi-channel video equipment at the same time, and the dense optical flow can be performed on the region of interest calculated from the dense optical flow in the video information at the same time.
  • the calculation can accurately determine the motion vector direction and motion vector distance of each frame of image, which provides an important premise guarantee for the subsequent determination of vehicles suspected to have exit/entry events in the berth area. For the vehicle entering the event, a large number of unnecessary calculation video frame images are filtered, which greatly reduces the calculation amount of the subsequent video image analysis to determine whether the vehicle has an exit/entry event, and also greatly reduces the vehicle.
  • the application example of the present invention aims to greatly improve the utilization rate of the computing power of the host under the premise of ensuring accurate analysis of parking events, thereby realizing efficient and accurate analysis and processing of parking events of multi-pole and multi-channel video cameras.
  • the roadside parking management system S includes a video array camera, an edge AI computing host C, a switch, a router, and a cloud server;
  • the video array camera is used to collect roadside parking video information, and many The video camera arrays in the pole position are connected to the same edge AI computing host C by cascading switches.
  • the switches are used for the edge AI computing host C to interconnect with each group of video arrays, and the routers are used for the edge AI computing host C and cloud servers.
  • the parking forensic data is uploaded between them; the cloud server is used to receive and process the parking forensic information and generate the parking forensic records.
  • the AI intelligent host C obtains the video image information of the monitoring area in real time, and determines the berth area in each video frame image; then expands the berth area in each video frame image to get The region of interest calculated by the dense optical flow of each video frame; the dense optical flow calculation is performed on the region of interest of each video frame, and the motion vector direction and motion vector distance of the region of interest of each frame image are obtained; The motion vector direction and motion vector distance of the region of interest in the image are calculated by dense optical flow in the statistical interval of a predetermined duration, the average value of the motion vector direction and the average motion vector distance of the region of interest in each frame of the image; according to the motion vector direction The average value and the average value of the motion vector distance are used to determine the vehicles suspected of having an exit/entry event in the parking area; finally, based on the video image information, the vehicle information of the vehicle suspected of having an exit/entry event is identified, and the vehicle information is determined Whether the vehicle has an exit
  • the edge AI computing host C in the embodiment of the present invention is composed of an embedded artificial intelligence computing platform, and the GPU of the platform includes 384 CUDA cores (a kind of arithmetic logic unit) and 48 Tensor cores (a new type of processing The core, it performs a special matrix mathematical operation, suitable for deep learning), and also contains 2 deep learning acceleration engines DLA (Data Link Address, data link address), with a total computing power of 21T, which is currently the highest computing power Edge-side embedded artificial intelligence platform.
  • DLA Data Link Address, data link address
  • step 101 before step 101 acquires video image information collected by multiple cameras in real time, and before the step of determining the parking area in each video frame image, the step includes: determining that each vertex of the vehicle parking space in the predetermined image collection area is at The coordinates in each image; according to the coordinates, determine the berth area in the video frame images captured by the video of each camera.
  • the coordinates of each vertex of the vehicle parking space in the predetermined image acquisition area in each image captured by the video array camera are determined, and then according to the coordinates of each vertex in each image, the video capture video of each camera is determined.
  • the berth area in the frame image; the determination of the berth area can be determined by debugging when the camera is installed on the pole, and there is no need to re-determine the real-time video image recognition. It should be noted that according to the needs of the actual scene, it can be manually Adjust and re-determine the berth area as needed, which is not limited here.
  • step 102 expands the berth area in each video frame image to obtain the area of interest calculated by the dense optical flow of each video frame, including: in the two-dimensional coordinate system of each video frame image, determining each The coordinate information of the berth area in the video frame image, and the long side of the berth area rectangle is extended for a predetermined length along the direction of the Y-axis value gradually decreasing to obtain the area of interest for the dense optical flow calculation of each video frame.
  • the dense optical flow calculation is performed on the region of interest to obtain the motion vector direction and motion vector distance of the region of interest in each frame of image, including: decoding each frame of image, and decoding each frame of images of interest in each frame of the image.
  • the dense optical flow calculation is performed on the region to obtain the motion vector direction and motion vector distance of the region of interest in each frame of image.
  • the parking area in the video frame image captured by the video of each camera is pre-determined; the video image information of the monitoring area is obtained in real time through multiple cameras of the video array camera, and the AI intelligent host C uses the rtsp protocol in real time.
  • the berth area in the video frame image is determined, The berth area in each video frame image collected in real time; then, in the two-dimensional coordinate system of each video frame image, the coordinate information of the berth area in each video frame image is determined, wherein, in the two-dimensional coordinate system of each video frame image, each video frame image The upper left vertex of the frame image is the origin of the two-dimensional coordinate system of the video frame image.
  • the direction in which the axis value gradually decreases extends a predetermined length, and combined with the closed area formed by the determined width of the berth area, the area of interest calculated by the dense optical flow of each video frame is obtained.
  • the predetermined length of the long side extension of the rectangular frame of the berth area can be calibrated on site, that is, a 1.6-meter pole is used to stand upright on the berth rectangular frame farther away in the camera shooting angle direction.
  • the AI intelligent host C decodes each frame of image, and performs dense optical flow calculation on the region of interest in each frame of the decoded image to obtain the motion vector direction and motion vector distance of the region of interest in each frame of image.
  • Host C uses the built-in multi-channel parallel hardware decoder to perform real-time hardware decoding on the acquired h264 (digital video compression format) stream data of each channel of video to obtain decoded image frames; AI intelligent host C is decoding the h264 code.
  • While streaming data also includes cyclically overwriting and buffering the undecoded video frame images in a predetermined storage location according to a predetermined buffer frequency; for example, according to the frequency of buffering once in 30 minutes, the undecoded h264 original video stream data is stored in the file format.
  • the form is cached in the local memory of the AI intelligent host C, and the content of the previous cache is cyclically overwritten when it is cached again; it should be noted that although the predetermined length of the extension is determined in a specific way in the embodiment of the present invention, it can also be based on the actual work. It is determined by experience and is not limited here.
  • the most suitable region of interest for calculation can be determined, thereby providing an important premise guarantee for subsequent image analysis and calculation. Further, determining the most suitable region of interest for calculation can also maximize the computing power of the intelligent host , thereby reducing unnecessary calculations.
  • dense optical flow calculation is performed on the region of interest in each frame of image after decoding to obtain the motion vector direction and motion vector distance of the region of interest in each frame of image, including: for each frame of image , divide the area of interest of the current frame after decoding into a predetermined number of calculation areas on average; perform dense optical flow calculation on each of the calculation areas to obtain the motion vector direction and motion vector distance of each calculation area; The motion vector direction and motion vector distance of the calculation area are averaged to obtain the motion vector direction and motion vector distance of the region of interest in the current frame.
  • the AI intelligent host C acquires the video image information of the monitoring area in real time through multiple cameras of the video array camera, and determines the parking area in each video frame image; then expands each video frame image Then, for each frame of image, divide the area of interest of the decoded current frame into a predetermined number of calculation areas, such as dividing the current video The area of interest of the frame is divided into multiple calculation areas by 4 ⁇ 4 pixel blocks, and the dense optical flow calculation is performed on the calculation area of each 4 ⁇ 4 pixel image block to obtain the motion vector direction and motion of each calculation area of the current video frame.
  • each video frame contains the motion vector (u, v) of each 4 ⁇ 4 pixel block calculation area in the ROI (Region Of Interest) area, where u represents the horizontal component of the vector, v
  • the vertical component of the representative vector then the motion vector direction and motion vector distance of each calculation area are averaged, and the motion vector (u, v) of each block in the optical flow vector frame is calculated to obtain the current frame area of interest.
  • the average motion vector (U', V'), the average distance of the motion vector of the region of interest in the current frame is the modulus of the vector (U', V'), and the motion direction of the motion vector of the region of interest in the current frame is the vector (U' , V') and the angle between the x-direction of the image.
  • the motion vector direction and motion vector distance of the current frame can be accurately obtained, thereby providing necessary preconditions for subsequent determination of a suspected vehicle that is suspected to have an exit event.
  • step 105 determines, according to the average value of the direction of the motion vector and the average value of the distance of the motion vector, the vehicle suspected of having an exit/entry event in the parking area, including: judging the distance of the motion vector Whether the average value is greater than the predetermined pixel threshold, if it is greater, it is determined that there is a vehicle suspected of having an entry and exit event in the parking area; in the two-dimensional coordinate system of any video frame image of each video image, determine the direction of the motion vector The movement direction of the average value; if the movement direction of the average value of the motion vector direction points to the parking area, it is determined that the vehicle is suspected to have an entry event; The vehicle appears to have been involved in an incident.
  • the dense optical flow is calculated within a statistical interval of a predetermined duration, such as 20 seconds It is a statistical interval, the average value of the motion vector direction and the average value of the motion vector distance of the region of interest of each frame image, and then, it is judged whether the average value of the motion vector distance is greater than the predetermined pixel threshold, if it is greater than, it is determined that there is a suspected occurrence in the berth area.
  • a predetermined duration such as 20 seconds
  • the vehicle entering the event then, in the two-dimensional coordinate system of any video frame image of each video image, determine the motion direction of the average value of the motion vector direction; if the motion direction of the average value of the motion vector direction points to the parking area , to determine that the suspected vehicle is suspected of having an admission incident.
  • the utilization ratio provides necessary preconditions for processing multi-channel video information at the same time, and avoids the occurrence of recognition errors due to a small number of frame errors, further improving the recognition accuracy.
  • step 106 identifies the vehicle information of the vehicle suspected of having an exit/entry event based on the video image information, and determines whether the vehicle has an exit/entry event according to the vehicle information, including: : Determine the start and end time of the suspected exit/entry event of the vehicle, and analyze the vehicle information of the vehicle in the video image information to be recognized within the start and end time; The vehicle performs detection and tracking to determine whether an exit/entry event occurs in the vehicle.
  • a video of the start and end times of the suspected exit/entry event of the vehicle is intercepted from the video information that has been cached in the predetermined storage location. segment, and then analyze the vehicle information of the vehicle in the video segment within the starting and ending time, including vehicle information such as license plate number, model, body color, etc., and detect and track the vehicle according to the vehicle information of the vehicle, so as to determine whether the vehicle has exited the scene. /entry event.
  • the relevant information of the exit/entry event of the vehicle is sent to the cloud server, and the cloud server generates parking forensics information, including vehicle close-up pictures, license plate numbers, entry and exit times, etc., and uploads the parking forensics information to the business platform.
  • the platform is closer to the parking permit information to complete the processing of the entire parking business.
  • the requirements for vehicle detection and tracking and the high real-time performance of the license plate algorithm are greatly reduced, and at the same time, based on the filtered video frame images, the vehicle entry and exit events can be determined efficiently and accurately, which greatly improves the parking lot. management efficiency.
  • Embodiments of the present invention provide an apparatus for managing on-street parking, which can implement the method embodiments provided above.
  • the method embodiments provided above For specific function implementation, please refer to the descriptions in the method embodiments, which will not be repeated here.
  • a general-purpose processor may be a microprocessor, or alternatively, the general-purpose processor may be any conventional processor, controller, microcontroller, or state machine.
  • a processor may also be implemented by a combination of computing devices, such as a digital signal processor and a microprocessor, multiple microprocessors, one or more microprocessors in combination with a digital signal processor core, or any other similar configuration. accomplish.
  • the steps of the method or algorithm described in the embodiments of the present invention may be directly embedded in hardware, a software module executed by a processor, or a combination of the two.
  • Software modules may be stored in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.
  • a storage medium may be coupled to the processor such that the processor may read information from, and store information in, the storage medium.
  • the storage medium can also be integrated into the processor.
  • the processor and storage medium may be provided in the ASIC, and the ASIC may be provided in the user terminal. Alternatively, the processor and the storage medium may also be provided in different components in the user terminal.
  • the above functions described in the embodiments of the present invention may be implemented in hardware, software, firmware, or any combination of the three. If implemented in software, the functions may be stored on, or transmitted over, a computer-readable medium in the form of one or more instructions or code.
  • Computer-readable media includes computer storage media and communication media that facilitate the transfer of a computer program from one place to another. Storage media can be any available media that a general-purpose or special-purpose computer can access.
  • Such computer-readable media may include, but are not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other device that can be used to carry or store instructions or data structures and Other media in the form of program code that can be read by a general-purpose or special-purpose computer, or a general-purpose or special-purpose processor.
  • any connection is properly defined as a computer-readable medium, for example, if software is transmitted from a web site, server or other remote source over a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL) Or transmitted by wireless means such as infrared, wireless, and microwave are also included in the definition of computer-readable media.
  • DSL digital subscriber line
  • the disks and disks include compact disks, laser disks, optical disks, DVDs, floppy disks and blu-ray disks. Disks usually reproduce data magnetically, while discs generally reproduce data optically with lasers. Combinations of the above can also be included in computer readable media.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Image Analysis (AREA)
  • Traffic Control Systems (AREA)

Abstract

一种路侧停车的管理方法及装置，管理方法包括：实时获取多个摄像机采集的视频图像信息，确定各视频帧图像中的泊位区域(S101)；扩展各视频帧图像中的泊位区域，得到各视频帧的稠密光流计算的感兴趣区域(S102)；对感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离(S103)；通过稠密光流计算在预定时长的统计区间内，各帧图像感兴趣区域的运动矢量方向平均值和运动矢量距离平均值(S104)；确定泊位区域内疑似发生出场/入场事件的车辆(S105)；确定车辆是否发生出场/入场事件(S106)。管理方法极大提升了主机算力的利用率，从而实现了高效、精确地对多杆位、多路视频相机的停车事件进行分析处理。

Description

一种路侧停车的管理方法及装置 技术领域
本发明涉及智能交通技术领域，尤其涉及一种路侧停车的管理方法及装置。
背景技术
随着时代的发展，城市路侧智慧停车管理作为城市智慧交通管理中的极其重要一环，开始进入一个快速发展和应用的时期。当前人工智能、物联网、云计算、大数据等新一代信息技术快速发展，以及边缘侧GPU(Central Processing Unit，中央处理器)、NPU(Neural-network Processing Units，网络处理器)等AI(Artificial Intelligence，人工智能)加速计算硬件平台的算力快速提升，为城市级路侧智慧停车与人工智能技术的深度融合，提供了重要的技术基础。
目前在部分城市已商用落地的路侧停车管理技术主要包含了地磁、低位路面视频桩、高位视频几种方式，高位视频由于其取证信息丰富、证据链完备、安装位置高不易损坏、场景适应性强、可靠性高，已成为各大城市优先发展的路侧停车管理方式。
当前基于高位视频的路侧停车管理方案在技术方案上主要有如下几种：
技术一：视频流实时检测跟踪方式
该方式通过不间断的对视频相机采集的每一帧实时图像进行车辆目标检测和跟踪，同时对每一帧进行车牌识别，从而获得停车事件相关信息和车辆相关信息。但是，通过该方式进行管理，停车管理设备需要持续不断的对视频相机的每一帧图像进行实时跟踪处理，然而停车行为本身是一个间歇性的行为，必然在某时间段内，车辆存在无出入场的行为，由于无法预知车辆何时发生出入场动作，因此该方式只能持续地对视频图像进行分析，但是视频图像分析的过程需要使用深度学习目标检测与跟踪算法，需要消耗边缘AI计算主机大量的算力，因此将无法并行处理较多的视频相机图像，导致该方式无法同时管理更多的泊位。
技术二：基于图像差分算法感知方式
该方式对视频相机采集的每一帧图像进行差分比对，当发现停车目标区域有超过图像差分阈值的变化，即对停车过程进行更深入的视频图像分析处理。该方式由于使用了差分 检测算法，差分算法的算力消耗小，避免了技术一视频流实时检测跟踪方式下，不间断地对视频相机每一帧的实时图像进行车辆目标检测和实时跟踪算力的消耗浪费，一定程度上提高了单台边缘AI计算主机可以并行处理的视频相机个数。但是，差分算法只能检测图像像素值的大小变化，无法确定物体运动方向，行人、过路车等干扰因素将导致大量的差分误判，以及对停车事件发生的时间区间估计产生较大的误差，从而对后续的视频图像分析造成严重影响。
技术三：使用雷达等距离传感器辅助探测，结合视频图像分析
在该方式下，需要在杆位上安装距离传感器(如雷达传感器、超声波传感器等)，当有车辆出现在停车目标区域，距离传感器探测到目标区域的距离变化，则通知监控设备进行抓拍，并识别车辆车牌信息，从而进行泊位的车辆管理。但是，该管理方式需要增加距离传感器，使得管理成本较高，且在雨雪等环境下，距离传感器测量得到的距离精度误差极大，抗干扰能力差，从而造成检测精度大大降低。
因此，急需一种能够高效、精确地检测泊位车辆，同时管理成本较低的路侧停车管理方法。
发明内容
本发明实施例提供一种路侧停车的管理方法及装置，在保证停车事件精确分析的前提下，极大提升主机算力的利用率，从而实现了高效、精确地对多杆位、多路视频相机的停车事件进行分析处理。
一方面，本发明实施例提供了一种路侧停车的管理的方法，包括：
实时获取多个摄像机采集的视频图像信息，确定各视频帧图像中的泊位区域；
扩展各视频帧图像中的泊位区域，得到各视频帧的稠密光流计算的感兴趣区域；
对所述感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离；
根据得到的每一帧图像感兴趣区域的运动矢量方向和运动矢量距离，通过稠密光流计算在预定时长的统计区间内，各帧图像感兴趣区域的运动矢量方向平均值和运动矢量距离平均值；
根据所述运动矢量方向平均值和运动矢量距离平均值，确定所述泊位区域内疑似发生出场/入场事件的车辆；
基于所述视频图像信息，识别疑似发生出场/入场事件车辆的车辆信息，并根据所述 车辆信息确定所述车辆是否发生出场/入场事件。
进一步地，在所述实时获取多个摄像机采集的视频图像信息，确定各视频帧图像中的泊位区域的步骤之前，包括：
确定预定图像采集区域内车辆泊位的各个顶点在各个图像中的坐标；
根据所述坐标，确定各个摄像机视频采集视频帧图像中的泊位区域。
进一步地，所述扩展各视频帧图像中的泊位区域，得到各视频帧的稠密光流计算的感兴趣区域，包括：
在各视频帧图像二维坐标系中，确定各视频帧图像中泊位区域的坐标信息，并将泊位区域矩形框的长边沿着Y轴值逐渐减小的方向延伸预定长度，得到各视频帧的稠密光流计算的感兴趣区域。
进一步地，所述对所述感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离，包括：
解码每一帧图像，并对解码后各帧图像中的感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离。
进一步地，所述对解码后各帧图像中的感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离，包括：
针对每一帧图像，将解码后当前帧的感兴趣区域平均分为预定个数的计算区域；
对每一个所述计算区域进行稠密光流计算，得到各个计算区域的运动矢量方向和运动矢量距离；
将所述各个计算区域的运动矢量方向和运动矢量距离求平均值，得到当前帧感兴趣区域的运动矢量方向和运动矢量距离。
进一步地，所述根据所述运动矢量方向平均值和运动矢量距离平均值，确定所述泊位区域内疑似发生出场/入场事件的车辆，包括：
判断所述运动矢量距离平均值是否大于预定像素阈值，若大于，确定所述泊位区域内存在疑似发生出入场事件的车辆；
在各视频图像的任一视频帧图像的二维坐标系中，确定所述运动矢量方向平均值的运动方向；
若所述运动矢量方向平均值的运动方向指向泊位区域内，确定所述车辆疑似发生入场事件；
若所述运动矢量方向平均值运动方向指向泊位区域外，确定所述车辆疑似发生出场事 件。
进一步地，所述基于所述视频图像信息，识别疑似发生出场/入场事件车辆的车辆信息，并根据所述车辆信息确定所述车辆是否发生出场/入场事件，包括：
确定所述车辆疑似发生出场/入场事件的起止时间，分析所述起止时间内的待识别视频图像信息中车辆的车辆信息；
根据所述车辆信息，对所述待识别视频图像信息中的所述车辆进行检测跟踪，确定所述车辆是否发生出场/入场事件。
另一方面，本发明实施例提供了一种路侧停车的管理的装置，包括：
获取及确定模块，用于实时获取多个摄像机采集的视频图像信息，确定各视频帧图像中的泊位区域；
扩展模块，用于扩展各视频帧图像中的泊位区域，得到各视频帧的稠密光流计算的感兴趣区域；
第一计算模块，用于对所述感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离；
第二计算模块，用于根据得到的每一帧图像感兴趣区域的运动矢量方向和运动矢量距离，通过稠密光流计算在预定时长的统计区间内，各帧图像感兴趣区域的运动矢量方向平均值和运动矢量距离平均值；
第一确定模块，用于根据所述运动矢量方向平均值和运动矢量距离平均值，确定所述泊位区域内疑似发生出场/入场事件的车辆；
识别及确定模块，用于基于所述视频图像信息，识别疑似发生出场/入场事件车辆的车辆信息，并根据所述车辆信息确定所述车辆是否发生出场/入场事件。
进一步地，包括：
第二确定模块，用于确定预定图像采集区域内车辆泊位的各个顶点在各个图像中的坐标；
第三确定模块，用于根据所述坐标，确定各个摄像机视频采集视频帧图像中的泊位区域。
进一步地，所述扩展模块，具体用于
在各视频帧图像二维坐标系中，确定各视频帧图像中泊位区域的坐标信息，并将泊位区域矩形框的长边沿着Y轴值逐渐减小的方向延伸预定长度，得到各视频帧的稠密光流计算的感兴趣区域。
进一步地，所述第一计算模块，包括：
解码单元，用于解码每一帧图像，并对解码后各帧图像中的感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离。
进一步地，所述解码单元，具体用于
针对每一帧图像，将解码后当前帧的感兴趣区域平均分为预定个数的计算区域；
对每一个所述计算区域进行稠密光流计算，得到各个计算区域的运动矢量方向和运动矢量距离；
将所述各个计算区域的运动矢量方向和运动矢量距离求平均值，得到当前帧感兴趣区域的运动矢量方向和运动矢量距离。
进一步地，所述第一确定模块，具体用于
判断所述运动矢量距离平均值是否大于预定像素阈值，若大于，确定所述泊位区域内存在疑似发生出入场事件的车辆；
在各视频图像的任一视频帧图像的二维坐标系中，确定所述运动矢量方向平均值的运动方向；
若所述运动矢量方向平均值的运动方向指向泊位区域内，确定所述车辆疑似发生入场事件；
若所述运动矢量方向平均值运动方向指向泊位区域外，确定所述车辆疑似发生出场事件。
进一步地，所述识别及确定模块，具体用于
确定所述车辆疑似发生出场/入场事件的起止时间，分析所述起止时间内的待识别视频图像信息中车辆的车辆信息；
根据所述车辆信息，对所述待识别视频图像信息中的所述车辆进行检测跟踪，确定所述车辆是否发生出场/入场事件。
上述技术方案具有如下有益效果：通过本发明，能够同时对多路视频设备采集的视频信息进行图像分析计算，通过对视频信息中的稠密光流计算的感兴趣区域进行稠密光流计算，能够精确地确定每一帧图像的运动矢量方向和运动矢量距离，为后续确定泊位区域内疑似发生出场/入场事件的车辆提供了重要的前提保障，通过确定泊位区域内疑似发生出场/入场事件的车辆，过滤了大量的非必要计算的视频帧图像，极大地降低了后续进行视频图像分析确定该车辆是否发生出场/入场事件的计算量，同时，也极大地降低了对车辆检测跟踪以及车牌算法高实时性的要求，从而实现了高效、精确地对多杆位、多路视频相机的停 车事件进行分析处理；实现了在保证停车事件精确分析的前提下，极大地提升了主机算力的利用率，进一步地，极大地降低了停车管理的成本。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本发明实施例中路侧停车的管理方法流程图；
图2为本发明一实施例中路侧停车的管理装置结构示意图。
具体实施方式
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
本发明实施例上述技术方案具有如下有益效果：通过本发明，能够同时对多路视频设备采集的视频信息进行图像分析计算，通过对视频信息中的稠密光流计算的感兴趣区域进行稠密光流计算，能够精确地确定每一帧图像的运动矢量方向和运动矢量距离，为后续确定泊位区域内疑似发生出场/入场事件的车辆提供了重要的前提保障，通过确定泊位区域内疑似发生出场/入场事件的车辆，过滤了大量的非必要计算的视频帧图像，极大地降低了后续进行视频图像分析确定该车辆是否发生出场/入场事件的计算量，同时，也极大地降低了对车辆检测跟踪以及车牌算法高实时性的要求，从而实现了高效、精确地对多杆位、多路视频相机的停车事件进行分析处理；实现了在保证停车事件精确分析的前提下，极大地提升了主机算力的利用率，进一步地，极大地降低了停车管理的成本。
以下结合应用实例对本发明实施例上述技术方案进行详细说明：
本发明应用实例旨在保证停车事件精确分析的前提下，极大提升主机算力的利用率，从而实现了高效、精确地对多杆位、多路视频相机的停车事件进行分析处理。
在一种可能的实现方式中，如在路侧停车管理系统S中，包括视频阵列相机、边缘AI计算主机C、交换机、路由器和云端服务器；视频阵列相机用于采集路侧泊位视频信息， 多个杆位的视频相机阵列通过交换机级联的方式连接到同一台边缘AI计算主机C上，交换机用于边缘AI计算主机C与各组视频阵列互联，路由器用于边缘AI计算主机C与云服务器之间进行停车取证数据上传；云端服务器，用于接收、处理停车取证信息，生成停车取证记录。首先通过视频阵列相机的多个摄像机，如16个摄像机，AI智能主机C实时获取监控区域的视频图像信息，确定各视频帧图像中的泊位区域；随后扩展各视频帧图像中的泊位区域，得到各视频帧的稠密光流计算的感兴趣区域；对各视频帧的感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离；根据得到的每一帧图像感兴趣区域的运动矢量方向和运动矢量距离，通过稠密光流计算在预定时长的统计区间内，各帧图像感兴趣区域的运动矢量方向平均值和运动矢量距离平均值；根据该运动矢量方向平均值和运动矢量距离平均值，确定泊位区域内疑似发生出场/入场事件的车辆；最后，基于视频图像信息，识别疑似发生出场/入场事件车辆的车辆信息，并根据该车辆信息确定该车辆是否发生出场/入场事件。
本发明实施例中的边缘AI计算主机C，其核心组成是嵌入式人工智能计算平台，该平台的GPU包含了384个CUDA core(一种算术逻辑单元)和48个Tensor core(一种新型处理核心，它执行一种专门的矩阵数学运算，适用于深度学习)，另外还包含2个深度学习加速引擎DLA(Data Link Address，数据链路地址)，总算力达到21T，是目前算力最高的边缘侧嵌入式人工智能平台。需要说明的是，本发明实施例中虽会以特定边缘AI计算主机为例说明，但在此不做限定。
在一种可能的实现方式中，在步骤101实时获取多个摄像机采集的视频图像信息，确定各视频帧图像中的泊位区域的步骤之前，包括：确定预定图像采集区域内车辆泊位的各个顶点在各个图像中的坐标；根据所述坐标，确定各个摄像机视频采集视频帧图像中的泊位区域。
例如，在路侧停车管理系统S中，确定预定图像采集区域内车辆泊位的各个顶点在视频阵列相机采集的各个图像中的坐标，随后根据各图像中各顶点的坐标，确定各个摄像机视频采集视频帧图像中的泊位区域；其中，泊位区域的确定可以在立杆安装摄像机时，调试确定即可，实时进行视频图像识别时无需再次重新确定，需要说明的是，根据实际场景的需求，可人工按需调整重新确定泊位区域，在此不做限定。
在一种可能的实现方式中，步骤102扩展各视频帧图像中的泊位区域，得到各视频帧的稠密光流计算的感兴趣区域，包括：在各视频帧图像二维坐标系中，确定各视频帧图像中泊位区域的坐标信息，并将泊位区域矩形框的长边沿着Y轴值逐渐减小的方向延伸预定 长度，得到各视频帧的稠密光流计算的感兴趣区域。
其中，对所述感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离，包括：解码每一帧图像，并对解码后各帧图像中的感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离。
例如，在路侧停车管理系统S中，预先确定各个摄像机视频采集视频帧图像中的泊位区域；通过视频阵列相机的多个摄像机实时获取监控区域的视频图像信息，AI智能主机C通过rtsp协议实时获取视频阵列中的多个视频相机采集的h264码流数据，其中视频分辨率为1080*1920，帧率为15帧；随后，基于预先确定的各个摄像机视频采集视频帧图像中的泊位区域，确定实时采集的各视频帧图像中的泊位区域；随后在各视频帧图像二维坐标系中，确定各视频帧图像中泊位区域的坐标信息，其中，各视频帧图像二维坐标系中，各视频帧图像的左上角顶点为视频帧图像二维坐标系的原点，从原点向右是X轴的正方向，从原点向下是Y轴的正方向；随后将泊位区域矩形框的长边沿着Y轴值逐渐减小的方向延伸预定长度，结合已确定的泊位区域的宽度组成的封闭区域，得到各视频帧的稠密光流计算的感兴趣区域。其中，泊位区域矩形框的长边扩展的预定长度可以采用现场标定的方式，即采用一个1.6米的杆直立在摄像机拍摄角方向较远处的泊位矩形框较远处的短泊位线的两个顶点，同时记录杆体顶端在图像上的投影点，图像上泊位线顶点到杆体顶端投影点的距离即为扩展的长度。随后，AI智能主机C解码每一帧图像，并对解码后各帧图像中的感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离，其中AI智能主机C，采用自带的多路并行硬件解码器，对获取的各路视频的h264(数字视频压缩格式)码流数据进行实时硬件解码得到解码后的图像帧；AI智能主机C在解码h264码流数据的同时，还包括将根据预定的缓存频率将未解码的各视频帧图像循环覆盖缓存在预定存储位置；如根据30分钟缓存一次的频率，将未解码的h264原始视频流数据以文件的形式缓存在AI智能主机C本地的存储器上，再一次缓存时循环覆盖前一次缓存的内容；需要说明的是，本发明实施例中虽会以特定方式确定扩展的预定长度，也可根据实际工作经验进行确定，在此不做限定。
通过本实施例，能够确定最适合计算的感兴趣区域，从而为后续进行图像分析计算提供了重要的前提保障，进一步地，确定最适合计算的感兴趣区域也能够最大地发挥智能主机的计算能力，从而减少不必要的计算。
在一种可能的实现方式中，对解码后各帧图像中的感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离，包括：针对每一帧图像，将解 码后当前帧的感兴趣区域平均分为预定个数的计算区域；对每一个所述计算区域进行稠密光流计算，得到各个计算区域的运动矢量方向和运动矢量距离；将所述各个计算区域的运动矢量方向和运动矢量距离求平均值，得到当前帧感兴趣区域的运动矢量方向和运动矢量距离。
例如，在路侧停车管理系统S中，首先通过视频阵列相机的多个摄像机，AI智能主机C实时获取监控区域的视频图像信息，确定各视频帧图像中的泊位区域；随后扩展各视频帧图像中的泊位区域，得到各视频帧的稠密光流计算的感兴趣区域；随后，针对每一帧图像，将解码后当前帧的感兴趣区域平均分为预定个数的计算区域，如将当前视频帧的感兴趣区域按4×4像素块分成多个计算区域，对每个4×4像素的图像块的计算区域进行稠密光流计算，得到当前视频帧的各计算区域的运动矢量方向和运动矢量距离，其中各视频帧中包含了ROI(Region Of Interest，感兴趣区域)区域中每个4×4像素块计算区域的运动矢量(u,v)，其中，u代表矢量的水平分量，v代表矢量的垂直分量；随后将各计算区域的运动矢量方向和运动矢量距离求平均值，根据光流矢量帧的中每个块的运动矢量(u，v)，计算得到当前帧感兴趣区域的平均运动矢量(U’，V’)，当前帧感兴趣区域的运动矢量距离平均值为向量(U’，V’)的模，当前帧感兴趣区域的运动矢量的运动方向为向量(U’，V’)与图像x方向的夹角。
通过本实施例，能够精确地得到当前帧的运动矢量方向和运动矢量距离，从而为后续确定疑似发生出场事件的疑似车辆提供了必要的前提条件。
在一种可能的实现方式中，步骤105根据所述运动矢量方向平均值和运动矢量距离平均值，确定所述泊位区域内疑似发生出场/入场事件的车辆，包括：判断所述运动矢量距离平均值是否大于预定像素阈值，若大于，确定所述泊位区域内存在疑似发生出入场事件的车辆；在各视频图像的任一视频帧图像的二维坐标系中，确定所述运动矢量方向平均值的运动方向；若所述运动矢量方向平均值的运动方向指向泊位区域内，确定所述车辆疑似发生入场事件；若所述运动矢量方向平均值运动方向指向泊位区域外，确定所述车辆疑似发生出场事件。
例如，接上例，在路侧停车管理系统S中，根据已得到的每一帧运动矢量方向平均值和运动矢量距离平均值，通过稠密光流计算在预定时长的统计区间内，如20秒为一统计区间，各帧图像感兴趣区域的运动矢量方向平均值和运动矢量距离平均值，随后，判断该运动矢量距离平均值是否大于预定像素阈值，若大于，确定泊位区域内存在疑似发生出入场事件的车辆；随后，在各视频图像的任一视频帧图像的二维坐标系中，确定该运动矢量 方向平均值的运动方向；若该运动矢量方向平均值的运动方向指向泊位区域内，确定疑似车辆疑似发生入场事件。
通过本实施例，过滤了大量的非必要计算的视频帧图像，极大地降低了后续进行视频图像分析确定所述疑似车辆是否发生出场/入场事件的计算量，从而提高了智能主机的算力利用率，为能够同时处理多路视频信息提供了必要的前提条件，且避免了因少量帧的误差而导致识别错误的情况发生，进一步地提高了识别的准确率。
在一种可能的实现方式中，步骤106基于所述视频图像信息，识别疑似发生出场/入场事件车辆的车辆信息，并根据所述车辆信息确定所述车辆是否发生出场/入场事件，包括：确定所述车辆疑似发生出场/入场事件的起止时间，分析所述起止时间内的待识别视频图像信息中车辆的车辆信息；根据所述车辆信息，对所述待识别视频图像信息中的所述车辆进行检测跟踪，确定所述车辆是否发生出场/入场事件。
例如，接上例，在路侧停车管理系统S中，确定车辆疑似发生入场事件后，在已缓存在预定存储位置的视频信息中截取该车辆疑似发生出场/入场事件的起止时间的视频片段，随后分析起止时间内的视频片段中该车辆的车辆信息，包括车牌号码、车型、车身颜色等车辆信息，并根据该车辆的车辆信息对该车辆进行检测跟踪，从而确定该车辆是否发生出场/入场事件。随后将该车辆发生出场/入场事件的相关信息发送至云端服务器，云端服务器生成停车取证信息，包括车辆特写图片、车牌号码、出入场时间等，并将停车取证信息上传业务平台，使得业务平台更近停车取证信息完成整个停车业务的处理。
通过本实施例，极大地降低了对车辆检测跟踪以及车牌算法高实时性的要求，同时，基于过滤后的视频帧图像，能够高效、精确地确定车辆的出入场事件，极大地提高了停车管理的效率。
本发明实施例提供了一种路侧停车的管理的装置，可以实现上述提供的方法实施例，具体功能实现请参见方法实施例中的说明，在此不再赘述。
应该明白，公开的过程中的步骤的特定顺序或层次是示例性方法的实例。基于设计偏好，应该理解，过程中的步骤的特定顺序或层次可以在不脱离本公开的保护范围的情况下得到重新安排。所附的方法权利要求以示例性的顺序给出了各种步骤的要素，并且不是要限于所述的特定顺序或层次。
在上述的详细描述中，各种特征一起组合在单个的实施方案中，以简化本公开。不应该将这种公开方法解释为反映了这样的意图，即，所要求保护的主题的实施方案需要比清楚地在每个权利要求中所陈述的特征更多的特征。相反，如所附的权利要求书所反映的那 样，本发明处于比所公开的单个实施方案的全部特征少的状态。因此，所附的权利要求书特此清楚地被并入详细描述中，其中每项权利要求独自作为本发明单独的优选实施方案。
为使本领域内的任何技术人员能够实现或者使用本发明，上面对所公开实施例进行了描述。对于本领域技术人员来说；这些实施例的各种修改方式都是显而易见的，并且本文定义的一般原理也可以在不脱离本公开的精神和保护范围的基础上适用于其它实施例。因此，本公开并不限于本文给出的实施例，而是与本申请公开的原理和新颖性特征的最广范围相一致。
上文的描述包括一个或多个实施例的举例。当然，为了描述上述实施例而描述部件或方法的所有可能的结合是不可能的，但是本领域普通技术人员应该认识到，各个实施例可以做进一步的组合和排列。因此，本文中描述的实施例旨在涵盖落入所附权利要求书的保护范围内的所有这样的改变、修改和变型。此外，就说明书或权利要求书中使用的术语“包含”，该词的涵盖方式类似于术语“包括”，就如同“包括，”在权利要求中用作衔接词所解释的那样。此外，使用在权利要求书的说明书中的任何一个术语“或者”是要表示“非排它性的或者”。
本领域技术人员还可以了解到本发明实施例列出的各种说明性逻辑块(illustrative logical block)，单元，和步骤可以通过电子硬件、电脑软件，或两者的结合进行实现。为清楚展示硬件和软件的可替换性(interchangeability)，上述的各种说明性部件(illustrative components)，单元和步骤已经通用地描述了它们的功能。这样的功能是通过硬件还是软件来实现取决于特定的应用和整个系统的设计要求。本领域技术人员可以对于每种特定的应用，可以使用各种方法实现所述的功能，但这种实现不应被理解为超出本发明实施例保护的范围。
本发明实施例中所描述的各种说明性的逻辑块，或单元都可以通过通用处理器，数字信号处理器，专用集成电路(ASIC)，现场可编程门阵列或其它可编程逻辑装置，离散门或晶体管逻辑，离散硬件部件，或上述任何组合的设计来实现或操作所描述的功能。通用处理器可以为微处理器，可选地，该通用处理器也可以为任何传统的处理器、控制器、微控制器或状态机。处理器也可以通过计算装置的组合来实现，例如数字信号处理器和微处理器，多个微处理器，一个或多个微处理器联合一个数字信号处理器核，或任何其它类似的配置来实现。
本发明实施例中所描述的方法或算法的步骤可以直接嵌入硬件、处理器执行的软件模块、或者这两者的结合。软件模块可以存储于RAM存储器、闪存、ROM存储器、EPROM 存储器、EEPROM存储器、寄存器、硬盘、可移动磁盘、CD-ROM或本领域中其它任意形式的存储媒介中。示例性地，存储媒介可以与处理器连接，以使得处理器可以从存储媒介中读取信息，并可以向存储媒介存写信息。可选地，存储媒介还可以集成到处理器中。处理器和存储媒介可以设置于ASIC中，ASIC可以设置于用户终端中。可选地，处理器和存储媒介也可以设置于用户终端中的不同的部件中。
在一个或多个示例性的设计中，本发明实施例所描述的上述功能可以在硬件、软件、固件或这三者的任意组合来实现。如果在软件中实现，这些功能可以存储与电脑可读的媒介上，或以一个或多个指令或代码形式传输于电脑可读的媒介上。电脑可读媒介包括电脑存储媒介和便于使得让电脑程序从一个地方转移到其它地方的通信媒介。存储媒介可以是任何通用或特殊电脑可以接入访问的可用媒体。例如，这样的电脑可读媒体可以包括但不限于RAM、ROM、EEPROM、CD-ROM或其它光盘存储、磁盘存储或其它磁性存储装置，或其它任何可以用于承载或存储以指令或数据结构和其它可被通用或特殊电脑、或通用或特殊处理器读取形式的程序代码的媒介。此外，任何连接都可以被适当地定义为电脑可读媒介，例如，如果软件是从一个网站站点、服务器或其它远程资源通过一个同轴电缆、光纤电缆、双绞线、数字用户线(DSL)或以例如红外、无线和微波等无线方式传输的也被包含在所定义的电脑可读媒介中。所述的碟片(disk)和磁盘(disc)包括压缩磁盘、镭射盘、光盘、DVD、软盘和蓝光光盘，磁盘通常以磁性复制数据，而碟片通常以激光进行光学复制数据。上述的组合也可以包含在电脑可读媒介中。
以上所述的具体实施方式，对本发明的目的、技术方案和有益效果进行了进一步详细说明，所应理解的是，以上所述仅为本发明的具体实施方式而已，并不用于限定本发明的保护范围，凡在本发明的精神和原则之内，所做的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (14)

  1. 一种路侧停车的管理方法，其特征在于，包括：
    实时获取多个摄像机采集的视频图像信息，确定各视频帧图像中的泊位区域；
    扩展各视频帧图像中的泊位区域，得到各视频帧的稠密光流计算的感兴趣区域；
    对所述感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离；
    根据得到的每一帧图像感兴趣区域的运动矢量方向和运动矢量距离，通过稠密光流计算在预定时长的统计区间内，各帧图像感兴趣区域的运动矢量方向平均值和运动矢量距离平均值；
    根据所述运动矢量方向平均值和运动矢量距离平均值，确定所述泊位区域内疑似发生出场/入场事件的车辆；
    基于所述视频图像信息，识别疑似发生出场/入场事件车辆的车辆信息，并根据所述车辆信息确定所述车辆是否发生出场/入场事件。
  2. 根据权利要求1所述的方法，其特征在于，在所述实时获取多个摄像机采集的视频图像信息，确定各视频帧图像中的泊位区域的步骤之前，包括：
    确定预定图像采集区域内车辆泊位的各个顶点在各个图像中的坐标；
    根据所述坐标，确定各个摄像机视频采集视频帧图像中的泊位区域。
  3. 根据权利要求1或2所述的方法，其特征在于，所述扩展各视频帧图像中的泊位区域，得到各视频帧的稠密光流计算的感兴趣区域，包括：
    在各视频帧图像二维坐标系中，确定各视频帧图像中泊位区域的坐标信息，并将泊位区域矩形框的长边沿着Y轴值逐渐减小的方向延伸预定长度，得到各视频帧的稠密光流计算的感兴趣区域。
  4. 根据权利要求3所述的方法，其特征在于，所述对所述感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离，包括：
    解码每一帧图像，并对解码后各帧图像中的感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离。
  5. 根据权利要求4所述的方法，其特征在于，所述对解码后各帧图像中的感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离，包括：
    针对每一帧图像，将解码后当前帧的感兴趣区域平均分为预定个数的计算区域；
    对每一个所述计算区域进行稠密光流计算，得到各个计算区域的运动矢量方向和运动矢量距离；
    将所述各个计算区域的运动矢量方向和运动矢量距离求平均值，得到当前帧感兴趣区域的运动矢量方向和运动矢量距离。
  6. 根据权利要求5所述的方法，其特征在于，所述根据所述运动矢量方向平均值和运动矢量距离平均值，确定所述泊位区域内疑似发生出场/入场事件的车辆，包括：
    判断所述运动矢量距离平均值是否大于预定像素阈值，若大于，确定所述泊位区域内存在疑似发生出入场事件的车辆；
    在各视频图像的任一视频帧图像的二维坐标系中，确定所述运动矢量方向平均值的运动方向；
    若所述运动矢量方向平均值的运动方向指向泊位区域内，确定所述车辆疑似发生入场事件；
    若所述运动矢量方向平均值运动方向指向泊位区域外，确定所述车辆疑似发生出场事件。
  7. 根据权利要求6所述的方法，其特征在于，所述基于所述视频图像信息，识别疑似发生出场/入场事件车辆的车辆信息，并根据所述车辆信息确定所述车辆是否发生出场/入场事件，包括：
    确定所述车辆疑似发生出场/入场事件的起止时间，分析所述起止时间内的待识别视频图像信息中车辆的车辆信息；
    根据所述车辆信息，对所述待识别视频图像信息中的所述车辆进行检测跟踪，确定所述车辆是否发生出场/入场事件。
  8. 一种路侧停车的管理装置，其特征在于，包括：
    获取及确定模块，用于实时获取多个摄像机采集的视频图像信息，确定各视频帧图像中的泊位区域；
    扩展模块，用于扩展各视频帧图像中的泊位区域，得到各视频帧的稠密光流计算的感兴趣区域；
    第一计算模块，用于对所述感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离；
    第二计算模块，用于根据得到的每一帧图像感兴趣区域的运动矢量方向和运动矢量距离，通过稠密光流计算在预定时长的统计区间内，各帧图像感兴趣区域的运动矢量方向平 均值和运动矢量距离平均值；
    第一确定模块，用于根据所述运动矢量方向平均值和运动矢量距离平均值，确定所述泊位区域内疑似发生出场/入场事件的车辆；
    识别及确定模块，用于基于所述视频图像信息，识别疑似发生出场/入场事件车辆的车辆信息，并根据所述车辆信息确定所述车辆是否发生出场/入场事件。
  9. 根据权利要求8所述的装置，其特征在于，包括：
    第二确定模块，用于确定预定图像采集区域内车辆泊位的各个顶点在各个图像中的坐标；
    第三确定模块，用于根据所述坐标，确定各个摄像机视频采集视频帧图像中的泊位区域。
  10. 根据权利要求8或9所述的装置，其特征在于，所述扩展模块，具体用于
    在各视频帧图像二维坐标系中，确定各视频帧图像中泊位区域的坐标信息，并将泊位区域矩形框的长边沿着Y轴值逐渐减小的方向延伸预定长度，得到各视频帧的稠密光流计算的感兴趣区域。
  11. 根据权利要求10所述的装置，其特征在于，所述第一计算模块，包括：
    解码单元，用于解码每一帧图像，并对解码后各帧图像中的感兴趣区域进行稠密光流计算，得到每一帧图像感兴趣区域的运动矢量方向和运动矢量距离。
  12. 根据权利要求11所述的装置，其特征在于，所述解码单元，具体用于
    针对每一帧图像，将解码后当前帧的感兴趣区域平均分为预定个数的计算区域；
    对每一个所述计算区域进行稠密光流计算，得到各个计算区域的运动矢量方向和运动矢量距离；
    将所述各个计算区域的运动矢量方向和运动矢量距离求平均值，得到当前帧感兴趣区域的运动矢量方向和运动矢量距离。
  13. 根据权利要求12所述的装置，其特征在于，所述第一确定模块，具体用于
    判断所述运动矢量距离平均值是否大于预定像素阈值，若大于，确定所述泊位区域内存在疑似发生出入场事件的车辆；
    在各视频图像的任一视频帧图像的二维坐标系中，确定所述运动矢量方向平均值的运动方向；
    若所述运动矢量方向平均值的运动方向指向泊位区域内，确定所述车辆疑似发生入场事件；
    若所述运动矢量方向平均值运动方向指向泊位区域外，确定所述车辆疑似发生出场事件。
  14. 根据权利要求13所述的装置，其特征在于，所述识别及确定模块，具体用于
    确定所述车辆疑似发生出场/入场事件的起止时间，分析所述起止时间内的待识别视频图像信息中车辆的车辆信息；
    根据所述车辆信息，对所述待识别视频图像信息中的所述车辆进行检测跟踪，确定所述车辆是否发生出场/入场事件。
PCT/CN2021/112931 2021-03-23 2021-08-17 一种路侧停车的管理方法及装置 WO2022198897A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110317344.5 2021-03-23
CN202110317344.5A CN113066306B (zh) 2021-03-23 2021-03-23 一种路侧停车的管理方法及装置

Publications (1)

Publication Number Publication Date
WO2022198897A1 true WO2022198897A1 (zh) 2022-09-29

Family

ID=76561865

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/112931 WO2022198897A1 (zh) 2021-03-23 2021-08-17 一种路侧停车的管理方法及装置

Country Status (2)

Country Link
CN (1) CN113066306B (zh)
WO (1) WO2022198897A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113066306B (zh) * 2021-03-23 2022-07-08 超级视线科技有限公司 一种路侧停车的管理方法及装置
CN114038232B (zh) * 2021-10-28 2022-09-20 超级视线科技有限公司 一种基于边缘端算存结合的路侧停车管理方法及系统
CN114005068A (zh) * 2021-11-08 2022-02-01 支付宝(杭州)信息技术有限公司 监测货物移动的方法及装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104376741A (zh) * 2014-12-02 2015-02-25 深圳市捷顺科技实业股份有限公司 一种车位状态检测方法与系统
CN104658249A (zh) * 2013-11-22 2015-05-27 上海宝康电子控制工程有限公司 基于帧差和光流实现车辆快速检测的方法
CN109684996A (zh) * 2018-12-22 2019-04-26 北京工业大学 基于视频的实时车辆进出识别方法
WO2019095588A1 (zh) * 2017-11-16 2019-05-23 智慧互通科技有限公司 基于多摄像机的路侧停车管理方法、装置和系统
CN112037265B (zh) * 2020-11-04 2021-02-02 天津天瞳威势电子科技有限公司 一种库位跟踪方法及装置
CN113066306A (zh) * 2021-03-23 2021-07-02 超级视线科技有限公司 一种路侧停车的管理方法及装置

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9129524B2 (en) * 2012-03-29 2015-09-08 Xerox Corporation Method of determining parking lot occupancy from digital camera images
JP2014110028A (ja) * 2012-12-04 2014-06-12 Sony Corp 画像処理装置、画像処理方法およびプログラム
CN107730975B (zh) * 2017-09-13 2020-04-28 浙江大学 超市停车引导反向寻车和出场引导的系统和方法
US11100366B2 (en) * 2018-04-26 2021-08-24 Volvo Car Corporation Methods and systems for semi-automated image segmentation and annotation

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104658249A (zh) * 2013-11-22 2015-05-27 上海宝康电子控制工程有限公司 基于帧差和光流实现车辆快速检测的方法
CN104376741A (zh) * 2014-12-02 2015-02-25 深圳市捷顺科技实业股份有限公司 一种车位状态检测方法与系统
WO2019095588A1 (zh) * 2017-11-16 2019-05-23 智慧互通科技有限公司 基于多摄像机的路侧停车管理方法、装置和系统
CN109684996A (zh) * 2018-12-22 2019-04-26 北京工业大学 基于视频的实时车辆进出识别方法
CN112037265B (zh) * 2020-11-04 2021-02-02 天津天瞳威势电子科技有限公司 一种库位跟踪方法及装置
CN113066306A (zh) * 2021-03-23 2021-07-02 超级视线科技有限公司 一种路侧停车的管理方法及装置

Also Published As

Publication number Publication date
CN113066306B (zh) 2022-07-08
CN113066306A (zh) 2021-07-02

Similar Documents

Publication Publication Date Title
WO2022198897A1 (zh) 一种路侧停车的管理方法及装置
CN111368687B (zh) 一种基于目标检测和语义分割的人行道车辆违停检测方法
WO2018223955A1 (zh) 目标监控方法、目标监控装置、摄像机及计算机可读介质
JP2017033554A (ja) ビデオデータ分析方法、装置及び駐車場モニタリングシステム
CN104506804A (zh) 一种快速道路上的机动车异常行为监测装置及其方法
JP2017525064A (ja) ビデオデータを用いた活動モニタリングのためのシステム及び方法
WO2019183752A1 (zh) 一种车前积雪与结冰的检测报警方法、存储介质和服务器
Laureshyn et al. Application of automated video analysis for behavioural studies: concept and experience
CN113055823B (zh) 一种基于路侧停车管理共享单车的方法及装置
CN113033479B (zh) 一种基于多层感知的泊位事件识别方法及系统
JP7305965B2 (ja) ビデオ監視システムのパラメータ設定方法、装置及びビデオ監視システム
CN112861773A (zh) 一种基于多层次的泊位状态检测方法及系统
CN110580808A (zh) 一种信息处理方法、装置、电子设备及智能交通系统
JP6915219B2 (ja) コンピュータ実施方法、撮像システム、及び画像処理システム
CN113380021A (zh) 车辆状态检测方法、装置、服务器及计算机可读存储介质
CN113112813B (zh) 违章停车检测方法及装置
CN113221791A (zh) 车辆违停检测方法、装置、电子设备和存储介质
Yao et al. Embedded technology and algorithm for video-based vehicle queue length detection
US11917325B2 (en) Automated scope limiting for video analytics
CN115174889A (zh) 摄像机的位置偏移检测方法、电子设备和存储介质
CN113808408A (zh) 路侧停车检测方法、装置、设备及可读存储介质
CN114038232B (zh) 一种基于边缘端算存结合的路侧停车管理方法及系统
CN111444798B (zh) 电动自行车驾驶行为的识别方法、装置和计算机设备
KR102317311B1 (ko) 영상을 이용한 정보 분석 시스템 및 그에 관한 방법
Ge A spatial-temporal-map-based traffic video analytic model for large-scale cloud-based deployment

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: 21932509

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 13/02/2024)