Path planning method and system based on computer vision

Abstract

The invention relates to the technical field of path updating, and particularly discloses a path planning method and system based on computer vision, and the method comprises the steps: obtaining an equipment position in a motion scene, and determining the position and definition of a fixed camera according to the equipment position; determining the position of the edge end and the application range of the edge end according to the position of the fixed camera; a video shot by the fixed camera is acquired regularly based on the edge end, the video is identified based on the AI module, an abnormal target is positioned, and an abnormal position is determined; determining a motion path according to the equipment position, and correcting the motion path in real time according to the positioned abnormal position; visual information in a workshop is collected in real time based on a fixed camera, the visual information is recognized, the abnormal position is determined, the abnormal position serves as the position of the same level as the equipment position, and all the positions serve as the passing points of the path planning process, so that the motion path of the inspection robot is updated in real time, and the intelligent level is extremely high.

Description

Technical Field

[0001] The present invention relates to the technical field of path updating, and in particular to a path planning method and system based on computer vision. Background Art

[0002] In the Internet of Things (IoT) workshop, inspection robots are used to automatically perform tasks such as checking equipment status, monitoring environmental parameters, cleaning, etc. In order for the inspection robots to complete tasks efficiently and intelligently, path planning is one of their core functions. Path planning involves finding an optimal path for the robot from the starting point to the target, which usually needs to consider factors such as the layout of the workshop, obstacles, and task requirements.

[0003] The motion paths of existing inspection robots are all static paths, which will not be changed once determined. Even if they can be changed, they are changed by remotely receiving control instructions from staff. This is actually relatively disconnected from the entire IoT workshop. How to adjust the path of the inspection robot based on the data generated in the IoT workshop and improve its intelligence level is the technical problem that the technical solution of the present invention aims to solve. Summary of the invention

[0004] The purpose of the present invention is to provide a path planning method and system based on computer vision to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions: A path planning method based on computer vision, the method comprising: Obtaining the device position in the motion scene, and determining the position and clarity of the fixed camera according to the device position; The position of the edge end and its application scope are determined according to the position of the fixed camera; the application scope is a collection of fixed cameras connected to the edge end; the edge end has a built-in AI module; The edge end periodically obtains the video captured by the fixed camera, identifies the video based on the AI ​​module, locates the abnormal target, and determines the abnormal location; Determine the motion path based on the device location, and correct the motion path in real time based on the abnormal location found; Update the application scope of the edge end and the clarity of the fixed camera based on the located abnormal position.

[0006] As a further solution of the present invention: the step of obtaining the device position in the motion scene and determining the position and clarity of the fixed camera according to the device position includes: Query the equipment registration information of the sports scene, and query the equipment location and equipment level in the equipment registration information; Determine the impact range and the value of each point in the impact range according to the equipment level; With the device location as the center, count the impact range and the values ​​of each point in the impact range to create an impact area containing numerical values; Superimpose the influence areas of all devices to obtain the values ​​at each point in the motion scene; Evenly arrange fixed cameras in the motion scene, and determine the clarity of each fixed camera according to the value at each point in the motion scene; The calculation process of the value of each point in the motion scene is: ; ; In the formula, Indicates the position in the motion scene The value at Indicates Devices at location The value at is the total number of devices, By The central value determined by the device level of each device, is the preset correction factor, Indicates location With The distance between devices.

[0007] As a further solution of the present invention: the step of determining the position of the edge end and its application range according to the position of the fixed camera includes: Determine the resource proportion of fixed cameras based on their clarity; Clustering the fixed cameras according to the resource proportions; Calculate the cluster center of each type of fixed camera and select the cluster center as the edge position; Get the label of a type of fixed camera corresponding to the edge as the application scope of the edge.

[0008] As a further solution of the present invention: the step of acquiring the video captured by the fixed camera based on the edge end at a fixed time, identifying the video based on the AI ​​module, locating the abnormal target, and determining the abnormal position includes: Based on the timed reception of the video from the fixed camera at the edge, the video is converted into an image sequence; Subtract adjacent images in sequence to obtain a difference image; Calculating the data volume of the difference image, and selecting an image according to the data volume; The selected images are input into the AI ​​module to locate abnormal targets and determine the abnormal locations.

[0009] As a further solution of the present invention, the step of acquiring the video captured by the fixed camera based on the edge end at a fixed time, identifying the video based on the AI ​​module, locating the abnormal target, and determining the abnormal position also includes: Read the target located by the AI ​​module, extract the convolution features of the target, and construct a convolution feature table; wherein the convolution feature table contains the initial convolution features pre-entered by the staff; When the difference image is calculated, the pixels in the difference image are traversed, the pixels whose pixel values ​​are greater than a preset threshold are marked, and the adjacent marked pixels are connected to obtain the target area; Eliminate the target area whose area is smaller than the preset area threshold to obtain the target frame; Based on the preset convolution feature table, the content in the target frame is identified, the abnormal target is located, and the abnormal position is determined; When the recognition is successful, the AI ​​module is not activated.

[0010] As a further solution of the present invention: the step of updating the application range of the edge end and the clarity of the fixed camera according to the located abnormal position includes: For any fixed camera, read the located abnormal targets sequentially; Querying the risk value corresponding to the abnormal target in a preset value table, and adjusting the clarity of the fixed camera according to the risk value; Regularly update the resource ratio according to the clarity of each fixed camera and update the application scope of the edge end.

[0011] The technical solution of the present invention also provides a path planning system based on computer vision, the system comprising: A camera calibration module is used to obtain the device position in the motion scene and determine the position and clarity of the fixed camera according to the device position; An edge end calibration module, used to determine the position of the edge end and its application scope according to the position of the fixed camera; the application scope is a collection of fixed cameras connected to the edge end; the edge end has a built-in AI module; The video recognition module is used to obtain the video captured by the fixed camera based on the edge end, identify the video based on the AI ​​module, locate the abnormal target, and determine the abnormal position; The path update module is used to determine the movement path according to the device position and to correct the movement path in real time according to the abnormal position located; The recursive update module is used to update the application scope of the edge end and the clarity of the fixed camera according to the located abnormal position.

[0012] As a further solution of the present invention: the camera calibration module includes: A device information query unit, used to query the device registration information of the sports scene, and query the device location and device level in the device registration information; A range determination unit, used to determine the impact range and the value of each point in the impact range according to the device level; The influence area determination unit is used to take the device location as the center, count the influence range and the values ​​of each point in the influence range, and create an influence area containing numerical values; The influence area superposition unit is used to superimpose the influence areas of all devices to obtain the values ​​at each point in the motion scene; A camera arrangement unit, used to evenly arrange fixed cameras in the motion scene, and determine the clarity of each fixed camera according to the value at each point in the motion scene; The calculation process of the value of each point in the motion scene is: ; ; In the formula, Indicates the position in the motion scene The value at Indicates Devices at location The value at is the total number of devices, By The central value determined by the device level of each device, is the preset correction factor, Indicates location With The distance between devices.

[0013] As a further solution of the present invention: the edge end calibration module includes: A resource ratio determination unit, used to determine the resource ratio of the fixed camera according to the definition of the fixed camera; A clustering unit, used for clustering fixed cameras according to the resource proportion; A position determination unit, used for calculating the cluster center of each type of fixed camera, and selecting the cluster center as the position of the edge end; The range determination unit is used to obtain a label corresponding to a type of fixed camera at the edge end as the application range of the edge end.

[0014] As a further solution of the present invention: the video recognition module includes: A video conversion unit, configured to convert the video acquired by the fixed camera based on the timed reception of the video by the edge end into an image sequence; An image difference unit, used for sequentially difference adjacent images to obtain a difference image; A selection unit, used for calculating the data amount of the difference image and selecting the image according to the data amount; The AI ​​application unit is used to input the selected image into the AI ​​module, locate the abnormal target, and determine the abnormal position.

[0015] Compared with the prior art, the beneficial effects of the present invention are: the present invention collects visual information in the workshop in real time based on a fixed camera, identifies the visual information, determines the abnormal position, and regards the abnormal position as the same level position as the equipment position, and uses it as the waypoint in the path planning process, thereby updating the movement path of the inspection robot in real time, and the intelligence level is extremely high. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0017] Figure 1 This is a flowchart of the path planning method based on computer vision.

[0018] Figure 2 This is a structural block diagram of the path planning system based on computer vision. DETAILED DESCRIPTION

[0019] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] Figure 1 The flowchart of the path planning method based on computer vision is as follows. In an embodiment of the present invention, a path planning method based on computer vision includes: Step S100: obtaining a device position in a motion scene, and determining a position and definition of a fixed camera according to the device position; Motion scenes are scenes that require the configuration of robots, such as IoT workshops and unmanned work areas. In these scenes, there are generally multiple working devices. The device positions of the working devices in the motion scenes are obtained, and the position and clarity of the fixed camera are determined based on the device positions.

[0021] Among them, regarding the location of fixed cameras, an evenly arranged approach is generally adopted to conduct global monitoring of motion scenes, but the clarity is related to the working equipment. The closer to the working equipment, the higher the clarity required and the stronger the recognition ability. This differentiated approach can deploy monitoring resources to more important areas and improve cost utilization.

[0022] It should be noted that the angle of the fixed camera cannot be changed, it has high stability, and at the same cost, it omits the conversion function and has higher shooting performance.

[0023] Step S200: determining the position of the edge end and its application scope according to the position of the fixed camera; the application scope is a collection of fixed cameras connected to the edge end; the edge end has a built-in AI module; The edge end is a microprocessor installed in the motion scene with independent data processing capabilities. It is used to process the data collected by the fixed camera and feed the processing results back to the main end. This architecture is very efficient and is a specific application of edge computing.

[0024] Specifically, there is a data interaction channel between the edge end and the fixed camera. The position of the edge end can be determined according to the determined position of the fixed camera. The fixed camera corresponding to each edge end is the application range of the edge end; the application range can be composed of the identification of the fixed camera connected to the edge end, and the identification is generally a number.

[0025] Step S300: acquiring the video captured by the fixed camera based on the edge terminal, identifying the video based on the AI ​​module, locating the abnormal target, and determining the abnormal position; For each edge terminal, there is data interaction with some fixed cameras. It receives videos acquired by the fixed cameras at regular intervals, recognizes the videos based on AI, locates abnormal targets, and obtains their abnormal positions.

[0026] Step S400: determining a motion path according to the device position, and correcting the motion path in real time according to the abnormal position located; For the robot's motion process, a motion path can be determined according to the device position, such as the A-star algorithm. The A-star algorithm combines the advantages of the Best-First Search and Dijkstra algorithms to find the shortest path from the starting point to the target in the graph.

[0027] After the motion path is determined according to the device position, the located abnormal position is read, and the located abnormal position is used as a waypoint to update the motion path, thereby obtaining a real-time updated motion path.

[0028] It is worth mentioning that the process of obtaining the abnormal position is real-time. There are different abnormal positions at different times. There is such a situation that the abnormal position appears and disappears. When it appears, the motion path will include the abnormal position. When it disappears, the abnormal position will be eliminated and the motion path will no longer include the abnormal position.

[0029] Step S500: updating the application range of the edge end and the clarity of the fixed camera according to the located abnormal position; After the abnormal target is identified, the motion path is updated accordingly. This application has achieved the desired function. On this basis, considering the working pressure of each edge end, this application introduces an update process. For a certain edge end, the abnormal targets corresponding to each fixed camera are counted. The more abnormal targets there are, the shorter the video receiving cycle of the corresponding fixed camera will be, and the more videos will be obtained. Correspondingly, the number of fixed cameras connected to the edge end will decrease.

[0030] Regarding step S100, the step of obtaining the device position in the motion scene and determining the position and clarity of the fixed camera according to the device position includes: Query the equipment registration information of the sports scene, and query the equipment location and equipment level in the equipment registration information; Determine the impact range and the value of each point in the impact range according to the equipment level; With the device location as the center, count the impact range and the values ​​of each point in the impact range to create an impact area containing numerical values; Superimpose the influence areas of all devices to obtain the values ​​at each point in the motion scene; Fixed cameras are evenly arranged in the motion scene, and the clarity of each fixed camera is determined according to the value at each point in the motion scene.

[0031] In an example of the technical solution of the present invention, the equipment information in the motion scene will be pre-registered, including the equipment location and equipment level of each power equipment. By querying the equipment registration information of the motion scene, the desired data can be obtained; the equipment level represents the importance of the power equipment, and the influence range is determined according to the equipment level. The higher the equipment level, the larger the influence range. Combined with the equipment location, an influence area can be determined, which is called the influence zone. The influence zone is generally circular.

[0032] Superimpose the influence areas of all devices to obtain the values ​​of each point in the motion scene. The value at each point reflects the importance of the corresponding point. Based on this, obtain the monitoring range of the fixed camera. This application requires global monitoring of the entire motion scene. The set of monitoring ranges of all fixed cameras must be larger than the motion scene. The entire area of ​​the motion scene is divided by the monitoring range of the fixed camera to obtain sub-areas.

[0033] Since a numerical value representing the importance has been calculated for each point, the values ​​at each point in the sub-area can be accumulated to obtain a numerical sum, and the clarity of the fixed camera can be determined based on the numerical sum; the clarity of the fixed camera is proportional to the numerical sum; that is, the more important the area corresponding to the fixed camera is, the higher the clarity will be.

[0034] The calculation process of the value of each point in the motion scene is as follows: ; ; In the formula, Indicates the position in the motion scene The value at Indicates Devices at location The value at is the total number of devices, By The central value determined by the device level of each device, is the preset correction factor, Indicates location With The distance between devices.

[0035] The above calculation process actually calculates each influence area at the position The values ​​at the locations are then superimposed; where each influence area is at the location The value at is inversely proportional to the square of the distance. Of course, a Gaussian function can also be introduced to describe this inverse relationship, that is, the farther away from the center, the smaller the value, which is smoother.

[0036] Regarding step S200, the step of determining the position of the edge end and its application range according to the position of the fixed camera includes: Determine the resource proportion of fixed cameras based on their clarity; Clustering the fixed cameras according to the resource proportions; Calculate the cluster center of each type of fixed camera and select the cluster center as the edge position; Get the label of a type of fixed camera corresponding to the edge as the application scope of the edge.

[0037] The higher the clarity of the fixed camera, the higher the amount of resources required to process the corresponding video, and the higher the resource share. By clustering the fixed cameras according to the resource share, multiple categories of fixed cameras can be obtained. For a category of fixed cameras corresponding to a higher resource share, the performance of the selected edge must be higher. On the contrary, for a category of fixed cameras corresponding to a lower resource share, the performance of the selected edge must be lower. In order to minimize the transmission pressure of the data transmission process (between the fixed camera and the edge), the cluster center is selected as the location of the edge, and the label of the fixed camera of the corresponding category of the edge is obtained as the application scope of the edge.

[0038] Regarding step S400, the step of acquiring the video captured by the fixed camera based on the edge terminal at a fixed time, identifying the video based on the AI ​​module, locating the abnormal target, and determining the abnormal position includes: Based on the timed reception of the video from the fixed camera at the edge, the video is converted into an image sequence; Subtract adjacent images in sequence to obtain a difference image; Calculating the data volume of the difference image, and selecting an image according to the data volume; The selected images are input into the AI ​​module to locate abnormal targets and determine the abnormal locations.

[0039] In an example of the technical solution of the present invention, the target positioning process is specifically described. Based on the video acquired by the fixed camera received at the edge end at regular intervals, the video is converted into an image sequence (the audio can be eliminated), and adjacent images are subtracted in turn to obtain a difference image. For the surveillance video, only the pixels corresponding to the moving objects will have differences, and the other pixels are close to zero after subtraction. By calculating the data volume of the difference image, it can be determined whether there are large changes in the adjacent images. If there are large changes, it means that identification is required. At this time, the image with a data volume that reaches the preset data volume threshold is selected and input into the AI ​​module to locate the abnormal target and determine the abnormal position.

[0040] The existing AI modules have extremely strong recognition performance, with great recognition breadth and high recognition depth.

[0041] As a preferred embodiment of the technical solution of the present invention, the step of acquiring the video captured by the fixed camera based on the edge end at regular intervals, identifying the video based on the AI ​​module, locating the abnormal target, and determining the abnormal position also includes: Read the target located by the AI ​​module, extract the convolution features of the target, and construct a convolution feature table; wherein the convolution feature table contains the initial convolution features pre-entered by the staff; When the difference image is calculated, the pixels in the difference image are traversed, the pixels whose pixel values ​​are greater than a preset threshold are marked, and the adjacent marked pixels are connected to obtain the target area; Eliminate the target area whose area is smaller than the preset area threshold to obtain the target frame; Based on the preset convolution feature table, the content in the target frame is identified, the abnormal target is located, and the abnormal position is determined; When the recognition is successful, the AI ​​module is not activated.

[0042] In an example of the technical solution of the present invention, before applying the AI ​​module, a pre-local recognition process is introduced to locate the target of each video to obtain the target frame. The content in the target frame is first identified based on a preset convolution feature table, so that the abnormal target can be located and the abnormal position can be determined. This process is essentially an image traversal and comparison process and does not involve the AI ​​recognition process. The convolution kernel table is a preset table. The staff pre-acquires images under different abnormal targets, identifies the images, and then extracts image features, which are called convolution kernels.

[0043] The target recognition process based on the convolution feature table can be performed locally, with a faster recognition speed and less resource consumption. The disadvantage of the target recognition process based on the convolution feature table is that it is unable to identify abnormal targets that have not been counted in advance. At this time, after introducing the AI ​​recognition model, unknown targets can be identified, greatly improving the comprehensiveness of recognition.

[0044] Specifically, adjacent images are subtracted in turn to obtain a difference image. After the introduction of the previous local recognition process, the difference image needs to be processed, and the pixels in the difference image are traversed, and the pixels whose pixel values ​​are greater than a preset threshold are marked. The adjacent marked pixels are connected to obtain the target area, which is the active area where the moving object is located. The target area (noise) with a particularly small area is eliminated, and the remaining target area is used as the area of ​​the real moving object. The rectangle where the area is located is obtained, which is called the target box; the content in the target box is identified based on the preset convolution feature table. After successful matching, the abnormal target corresponding to the target box is obtained.

[0045] When the output of identifying the content in the target box based on the preset convolution feature table is empty, that is, when identifying the content in the target box based on the preset convolution feature table, the match fails. At this time, the AI ​​module in the edge is applied to identify the target. Each time the AI ​​module is applied, the convolution feature table will be updated once to improve the performance of the local recognition process.

[0046] It is worth mentioning that in actual applications, since there are only a limited number of abnormal targets, the convolutional feature table is sufficient to cover most types, and the application frequency of AI recognition models is actually very low.

[0047] Regarding step S500, the step of updating the application range of the edge end and the clarity of the fixed camera according to the located abnormal position includes: For any fixed camera, read the located abnormal targets sequentially; Querying the risk value corresponding to the abnormal target in a preset value table, and adjusting the clarity of the fixed camera according to the risk value; Regularly update the resource ratio according to the clarity of each fixed camera and update the application scope of the edge end.

[0048] In an example of the technical solution of the present invention, a negative feedback adjustment scheme is provided. For each fixed camera, the abnormal targets identified by it are read in sequence. A staff member pre-sets a numerical value indicating the degree of risk for each abnormal target, that is, a risk value. The risk value corresponding to the abnormal target is queried, and the video receiving cycle of the fixed camera is adjusted according to the risk value. The video receiving cycle is inversely proportional to the risk value. The larger the risk value, the shorter the video receiving cycle. Furthermore, a correction rate is determined according to the risk value, and the resource ratio of the fixed camera is adjusted. This affects the clustering process of the fixed camera (step S200), that is, which edge ends manage different fixed cameras. The performance of different edge ends is different. The higher the risk value, the higher the corresponding resource ratio.

[0049] Figure 2 : is a structural block diagram of a path planning system based on computer vision. In an embodiment of the present invention, a path planning system based on computer vision, the system 10 includes: A camera calibration module 11 is used to obtain the device position in the motion scene and determine the position and clarity of the fixed camera according to the device position; The edge end calibration module 12 is used to determine the position of the edge end and its application range according to the position of the fixed camera; the application range is a collection of fixed cameras connected to the edge end; the edge end has a built-in AI module; The video recognition module 13 is used to obtain the video captured by the fixed camera based on the edge terminal timing, identify the video based on the AI ​​module, locate the abnormal target, and determine the abnormal position; A path updating module 14 is used to determine the movement path according to the device position and to correct the movement path in real time according to the abnormal position located; The recursive updating module 15 is used to update the application scope of the edge end and the clarity of the fixed camera according to the located abnormal position.

[0050] Furthermore, the camera calibration module 11 includes: A device information query unit, used to query the device registration information of the sports scene, and query the device location and device level in the device registration information; A range determination unit, used to determine the impact range and the value of each point in the impact range according to the device level; The influence area determination unit is used to take the device location as the center, count the influence range and the values ​​of each point in the influence range, and create an influence area containing numerical values; The influence area superposition unit is used to superimpose the influence areas of all devices to obtain the values ​​at each point in the motion scene; A camera arrangement unit, used to evenly arrange fixed cameras in the motion scene, and determine the clarity of each fixed camera according to the value at each point in the motion scene; The calculation process of the value of each point in the motion scene is: ; ; In the formula, Indicates the position in the motion scene The value at Indicates Devices at location The value at is the total number of devices, By The central value determined by the device level of each device, is the preset correction factor, Indicates location With The distance between devices.

[0051] Specifically, the edge end calibration module 12 includes: A resource ratio determination unit, used to determine the resource ratio of the fixed camera according to the definition of the fixed camera; A clustering unit, used for clustering fixed cameras according to the resource proportion; A position determination unit, used for calculating the cluster center of each type of fixed camera, and selecting the cluster center as the position of the edge end; The range determination unit is used to obtain a label corresponding to a type of fixed camera at the edge end as the application range of the edge end.

[0052] Furthermore, the video recognition module 13 includes: A video conversion unit, configured to convert the video acquired by the fixed camera based on the timed reception of the video by the edge end into an image sequence; An image difference unit, used for sequentially difference adjacent images to obtain a difference image; A selection unit, used for calculating the data amount of the difference image and selecting the image according to the data amount; The AI ​​application unit is used to input the selected image into the AI ​​module, locate the abnormal target, and determine the abnormal position.

[0053] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A path planning method based on computer vision, characterized in that: The method comprises: Obtaining the device position in the motion scene, and determining the position and clarity of the fixed camera according to the device position; The position of the edge end and its application scope are determined according to the position of the fixed camera; the application scope is a collection of fixed cameras connected to the edge end; the edge end has a built-in AI module; The edge end periodically obtains the video captured by the fixed camera, identifies the video based on the AI ​​module, locates the abnormal target, and determines the abnormal location; Determine the motion path based on the device location, and correct the motion path in real time based on the abnormal location found; Update the application scope of the edge end and the clarity of the fixed camera based on the located abnormal position.

2. The computer vision-based path planning method according to claim 1, characterized in that: The step of obtaining the device position in the motion scene and determining the position and clarity of the fixed camera according to the device position comprises: Query the equipment registration information of the sports scene, and query the equipment location and equipment level in the equipment registration information; Determine the impact range and the value of each point in the impact range according to the equipment level; With the device location as the center, count the impact range and the values ​​of each point in the impact range to create an impact area containing numerical values; Superimpose the influence areas of all devices to obtain the values ​​at each point in the motion scene; Evenly arrange fixed cameras in the motion scene, and determine the clarity of each fixed camera according to the value at each point in the motion scene; The calculation process of the value of each point in the motion scene is: ; ; In the formula, Indicates the position in the motion scene The value at Indicates Devices at location The value at is the total number of devices, By The central value determined by the device level of each device, is the preset correction factor, Indicates location With The distance between devices.

3. The computer vision-based path planning method according to claim 1, characterized in that: The step of determining the position of the edge end and its application range according to the position of the fixed camera comprises: Determine the resource proportion of fixed cameras based on their clarity; Clustering the fixed cameras according to the resource proportions; Calculate the cluster center of each type of fixed camera and select the cluster center as the edge position; Get the label of a type of fixed camera corresponding to the edge as the application scope of the edge.

4. The computer vision-based path planning method according to claim 1, characterized in that: The step of acquiring the video captured by the fixed camera based on the edge end at a fixed time, identifying the video based on the AI ​​module, locating the abnormal target, and determining the abnormal position includes: Based on the timed reception of the video from the fixed camera at the edge, the video is converted into an image sequence; Subtract adjacent images in sequence to obtain a difference image; Calculating the data volume of the difference image, and selecting an image according to the data volume; The selected images are input into the AI ​​module to locate abnormal targets and determine the abnormal locations.

5. The computer vision-based path planning method according to claim 4, characterized in that: The step of acquiring the video captured by the fixed camera based on the edge end at a fixed time, identifying the video based on the AI ​​module, locating the abnormal target, and determining the abnormal position also includes: Read the target located by the AI ​​module, extract the convolution features of the target, and construct a convolution feature table; wherein the convolution feature table contains the initial convolution features pre-entered by the staff; When the difference image is calculated, the pixels in the difference image are traversed, the pixels whose pixel values ​​are greater than a preset threshold are marked, and the adjacent marked pixels are connected to obtain the target area; Eliminate the target area whose area is smaller than the preset area threshold to obtain the target frame; Based on the preset convolution feature table, the content in the target frame is identified, the abnormal target is located, and the abnormal position is determined; When the recognition is successful, the AI ​​module is not activated.

6. The computer vision-based path planning method according to claim 1, characterized in that: The step of updating the application scope of the edge end and the clarity of the fixed camera according to the located abnormal position includes: For any fixed camera, read the located abnormal targets sequentially; Querying the risk value corresponding to the abnormal target in a preset value table, and adjusting the clarity of the fixed camera according to the risk value; Regularly update the resource ratio according to the clarity of each fixed camera and update the application scope of the edge end.

7. A path planning system based on computer vision, characterized in that: The system comprises: A camera calibration module is used to obtain the device position in the motion scene and determine the position and clarity of the fixed camera according to the device position; An edge end calibration module, used to determine the position of the edge end and its application scope according to the position of the fixed camera; the application scope is a collection of fixed cameras connected to the edge end; the edge end has a built-in AI module; The video recognition module is used to obtain the video captured by the fixed camera based on the edge end, identify the video based on the AI ​​module, locate the abnormal target, and determine the abnormal position; A path update module is used to determine the movement path according to the device position and to correct the movement path in real time according to the abnormal position located; The recursive update module is used to update the application scope of the edge end and the clarity of the fixed camera according to the located abnormal position.

8. The computer vision-based path planning system according to claim 7, characterized in that: The camera calibration module comprises: A device information query unit, used to query the device registration information of the sports scene, and query the device location and device level in the device registration information; A range determination unit, used to determine the impact range and the value of each point in the impact range according to the device level; The influence area determination unit is used to count the influence range and the values ​​of each point in the influence range with the device location as the center, and create an influence area containing numerical values; The influence area superposition unit is used to superimpose the influence areas of all devices to obtain the values ​​at each point in the motion scene; A camera arrangement unit, used to evenly arrange fixed cameras in the motion scene, and determine the clarity of each fixed camera according to the value at each point in the motion scene; The calculation process of the value of each point in the motion scene is: ; ; In the formula, Indicates the position in the motion scene The value at Indicates Devices at location The value at is the total number of devices, By The central value determined by the device level of each device, is the preset correction factor, Indicates location With The distance between devices.

9. The computer vision-based path planning system according to claim 7, characterized in that: The edge terminal calibration module includes: A resource ratio determination unit, used to determine the resource ratio of the fixed camera according to the definition of the fixed camera; A clustering unit, used for clustering fixed cameras according to the resource proportion; A position determination unit, used to calculate the cluster center of each type of fixed camera, and select the cluster center as the position of the edge end; The range determination unit is used to obtain a label corresponding to a type of fixed camera at the edge end as the application range of the edge end.

10. The computer vision-based path planning system according to claim 7, characterized in that: The video recognition module comprises: A video conversion unit, configured to convert the video acquired by the fixed camera based on the timed reception of the video by the edge end into an image sequence; An image difference unit, used for sequentially difference adjacent images to obtain a difference image; A selection unit, used for calculating the data amount of the difference image and selecting an image according to the data amount; The AI ​​application unit is used to input the selected image into the AI ​​module, locate the abnormal target, and determine the abnormal position.

Images