Intelligent blind guiding method, wheelchair, computer equipment and storage medium

Abstract

The invention discloses an intelligent blind guiding method, a wheelchair, computer equipment and a storage medium, and belongs to the technical field of blind guiding. Comprising the following steps: receiving voice input of a user, and analyzing geographic position key information; sending a voice feedback instruction; the voice feedback instruction is used for interacting with a user to confirm whether the geographic position key information is destination information; if so, executing the following steps: setting an initial position point and a target point; generating N travel routes based on the initial position point and the target point; selecting an optimal route, and sending a navigation travel instruction based on the optimal route; a road condition image is acquired in real time, whether an obstacle exists in the road condition is judged based on the road condition image, and if yes, an obstacle avoidance instruction is sent; otherwise, a current situation keeping instruction is sent. According to the method, an original target detection loss function is replaced on the basis of the yolov8, and the Concenter-iou is adopted as a detection frame loss function, so that the obstacle detection precision of the model is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of blind guiding, and in particular relates to an intelligent blind guiding method, a wheelchair, a computer device and a storage medium. Background Art

[0002] The aging society has an increasing demand for travel assistance devices. Although traditional travel assistance methods such as manual care, blind path guidance, cane navigation and guide dogs have provided help to the visually impaired to a certain extent, they all have obvious limitations, such as strong dependence, limited coverage and high maintenance costs.

[0003] The emerging electronic guide system combines ultrasonic sensors, lidar, GPS and other sensing technologies to provide users with sound, vibration or other forms of navigation information through wearable devices or handheld devices. Lidar emits laser pulses and receives reflected signals to accurately measure the distance and shape of surrounding objects and generate a three-dimensional map of the environment.

[0004] Although the electronic guide system can provide more dimensional environmental information, it is sensitive to environmental conditions, has high system complexity, high user learning costs, and high maintenance and update costs. Although LiDAR technology can provide accurate three-dimensional environmental information, its high cost, unstable performance under special weather conditions, and difficulty in identifying the type and attributes of obstacles limit its application in the field of guides. Summary of the invention

[0005] Purpose of the invention: In order to solve the above problems, the present invention provides an intelligent blind guiding method, a wheelchair, a computer device and a storage medium.

[0006] Technical solution: An intelligent blind-guiding method, comprising the following steps:

[0007] Receive a user's voice input, and parse out key geographical location information based on the voice input; send a voice feedback instruction based on the key geographical location information; the voice feedback instruction is used to interact with the user to confirm whether the key geographical location information is information about the destination; if so, perform the following steps:

[0008] Acquire current location information and set it as an initial location point, and set the geographical location key information as a target point; generate N travel routes based on the initial location point and the target point; select an optimal route from the N travel routes, and send a navigation travel instruction based on the optimal route;

[0009] During the navigation process, the road condition image is obtained in real time, and whether there is an obstacle in the road condition is determined based on the road condition image: if so, an obstacle avoidance instruction is sent; otherwise, a maintain status quo instruction is sent, and the maintain status quo instruction is to continue moving and continuously obtain and analyze the road condition image until the target point is reached.

[0010] In a further embodiment, selecting the optimal route from N travel routes comprises the following steps:

[0011] Obtaining route optimization parameters for each travel route; the route optimization parameters at least include: route time, travel expenses, number of transfers, and passenger flow;

[0012] The route preference parameters are scored and assigned corresponding weight parameters; based on the scores and weight parameters, a preference score for each travel route is calculated; and based on the preference scores, an optimal route is determined from the N travel routes.

[0013] In a further embodiment, judging whether there is an obstacle in the road condition based on the road condition image includes the following steps:

[0014] The road condition image is preprocessed, the preprocessed road condition image is input into an obstacle recognition model, and obstacle information is output; the obstacle information at least includes object category information, object detection frame information, object category confidence, and the distance between the object and itself.

[0015] In a further embodiment, the obstacle recognition model is constructed as follows:

[0016] Obtaining obstacle images existing on the road, and annotating the obstacle images; dividing the annotated obstacle images into training set images and test set images; enhancing the training set images, and inputting the test set images into a target detection model to obtain a prediction frame;

[0017] A comparison is performed based on the position of the predicted frame and the position of the real frame annotated in the obstacle image, and a loss calculation is performed according to the comparison result; based on the loss calculation result, a target detection model is trained to obtain the obstacle recognition model.

[0018] In a further embodiment, the loss calculation comprises the following steps:

[0019] Calculate the angle loss ∧,

[0020]

[0021] Where N is the distance between detection boxes, c his the height of the rectangle with L as the diagonal; a is the angle between the width and the diagonal, and b is the angle between the height and the diagonal; is the center coordinate of the real frame, is the center coordinate of the prediction box;

[0022] Calculate the distance loss Δ:

[0023]

[0024] Where x and y are the horizontal and vertical coordinates of the center point of the detection frame respectively; c ω is the width of the rectangle with L as the diagonal; γ is a constant;

[0025] Calculate the shape loss Ω:

[0026]

[0027] Where ω is the width of the prediction box, ω gt is the width of the true box and h is the height of the predicted box. gt is the height of the real box; θ is a constant;

[0028] Finally, we calculate L SIoU :

[0029] Calculate L Concentrer―IoU :L Concentrer―IoU =1-IoU Concentrer ;

[0030]

[0031] Among them, A is the intersection of the real box and the predicted box, and B is the union of the real box and the predicted box;

[0032] L Concentrer―SIoU =L SIoU +IoU―IoU Concentrer ;

[0033] L 检测框 =L Concentrer―SIoU ;

[0034] L=L 检测框 +L 分类损失 .

[0035] In a further embodiment, preprocessing the road condition image comprises the following steps:

[0036] Scaling the road condition image, changing the channel order of the road condition image, converting the road condition image into a CHW format, and then expanding the dimension of the road condition image and converting it into a four-dimensional array BCHW;

[0037] The storage mode of the traffic condition images in the memory is changed to continuous storage, and the precision of the traffic condition images is converted to single precision;

[0038] The road condition image is normalized and input into the obstacle recognition model. The output information of the obstacle recognition model is subjected to non-maximum suppression processing, and the obstacle coordinates and the distance between the obstacle and itself are output.

[0039] In a further embodiment, the obstacle avoidance instruction includes the following process:

[0040] Calculate the priority of each obstacle based on the object category information of the obstacle and the distance between the obstacle and itself;

[0041] Based on the priorities, avoid obstacles in order;

[0042] For obstacles that are closer and have higher priority, the obstacle avoidance module will change the direction or speed of the wheelchair through the motor control system to bypass the obstacle;

[0043] The vehicle stops to avoid dynamic obstacles, while the vehicle actively bypasses static obstacles.

[0044] In another technical solution, an intelligent blind-guiding balancing wheelchair is provided, which is used to implement the intelligent blind-guiding method as described above, including:

[0045] The wheelchair body is used to move on the travel route;

[0046] The voice module is used to receive the user's voice input and send voice feedback instructions;

[0047] A route planning module is used to generate N travel routes based on the initial location point and the target point;

[0048] The visual module is used to obtain real-time road condition images during navigation;

[0049] An obstacle avoidance module, used for determining whether there is an obstacle in the road condition based on the road condition image;

[0050] A balance sensing module is used to sense whether the wheelchair is in a balanced state;

[0051] A motor control module is used to keep the wheelchair body in a balanced state;

[0052] The central processing module is used to connect, manage and control the voice module, route planning module, vision module, obstacle avoidance module, balance perception module and motor control module.

[0053] In another technical solution, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an intelligent blind guiding method as described above is implemented.

[0054] In another technical solution, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the intelligent blind guiding method as described above is implemented.

[0055] Beneficial effects:

[0056] (1) Based on yolov8, the original target detection loss function is replaced and concentrer-iou is used as the detection box loss function, which effectively improves the model's detection accuracy for obstacles;

[0057] (2) The guide wheelchair can efficiently identify and avoid obstacles in the environment, solving the limitations that traditional assistive tools such as manual care, blind path guidance, cane navigation and guide dogs cannot overcome;

[0058] (3) The guide wheelchair achieves automatic control of seat balance, reduces the danger caused by seat tilt, and improves the user's comfort. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a work flow chart of the present invention;

[0060] Figure 2 is a topological diagram of each module in the wheelchair of the present invention;

[0061] Figure 3 This is the workflow diagram of the obstacle avoidance module;

[0062] Figure 4 is the definition diagram of angle loss;

[0063] Figure 5 is the definition graph of distance loss;

[0064] Figure 6 This is the workflow diagram of the balance perception module. DETAILED DESCRIPTION

[0065] Example 1

[0066] like Figures 1 to 6 As shown, this embodiment provides an intelligent blind guiding method, comprising the following steps:

[0067] Step 1: Receive a user's voice input, and parse out key geographic location information based on the voice input; send a voice feedback instruction based on the key geographic location information; the voice feedback instruction is used to interact with the user to confirm whether the key geographic location information is the destination information:

[0068] If yes, perform the following steps:

[0069] Step 2: Obtain the current location information and set it as the initial location point, and set the geographical location key information as the target point; generate N travel routes based on the initial location point and the target point; select the optimal route from the N travel routes, and send a navigation travel instruction based on the optimal route.

[0070] Further, selecting the optimal route from the N travel routes includes the following steps:

[0071] Obtaining route optimization parameters for each travel route; the route optimization parameters at least include: route time, travel expenses, number of transfers, and passenger flow;

[0072] The route preference parameters are scored and assigned corresponding weight parameters; based on the scores and weight parameters, a preference score for each travel route is calculated; and based on the preference scores, an optimal route is determined from the N travel routes.

[0073] The preferred score of each travel route = path preferred parameter score * weight parameter. For example, the preferred score of each travel route = highest priority parameter score * 0.5 + second priority parameter score * 0.25 + third priority parameter score * 0.125 + fourth priority parameter * 0.0625. In addition, based on actual needs, different priorities are set for route time, distance cost, number of transfers, and passenger flow. For example, if route time is set as the highest priority parameter, the score is 80; if the number of transfers is set as the second priority parameter, the score is 85; if the passenger flow is set as the third priority parameter, the score is 65; if the distance cost is set as the fourth priority parameter, the score is 75. From the N travel routes, the travel route with the highest preferred score is determined as the optimal route.

[0074] Step 3: During the navigation process, a road condition image is obtained in real time, and based on the road condition image, it is determined whether there are obstacles in the road condition: if so, an obstacle avoidance instruction is sent; otherwise, a maintain status quo instruction is sent, and the maintain status quo instruction is to continue moving and continuously obtain and analyze road condition images until the target point is reached.

[0075] Further, judging whether there is an obstacle in the road condition based on the road condition image includes the following steps:

[0076] The road condition image is preprocessed, the preprocessed road condition image is input into an obstacle recognition model, and the coordinates and distance of the obstacle are output; the output information includes object category information, object detection frame information, object category confidence, and the distance between the object and itself.

[0077] Furthermore, the construction process of the obstacle recognition model is as follows:

[0078] Obtaining obstacle images existing on the road, and annotating the obstacle images; dividing the annotated obstacle images into training set images and test set images; enhancing the training set images, and inputting the test set images into a target detection model to obtain a prediction frame;

[0079] A comparison is performed based on the position of the predicted frame and the position of the real frame annotated in the obstacle image, and a loss calculation is performed according to the comparison result; based on the loss calculation result, a target detection model is trained to obtain the obstacle recognition model.

[0080] The method for making the model is as follows: ① First, collect obstacle images on the road.

[0081] ② First, organize the obstacle image data on the road and divide the obstacle image data into training set and test set according to the ratio of 8:2.

[0082] ③ Use annotation software to annotate data. The real frame information includes the coordinates of the center point of the real frame and the length and width of the real frame;

[0083] ④ The labeled data is divided into training set and test set according to the set ratio.

[0084] ⑤ Perform Cutout data enhancement on the training set images: first randomly read images from the data set, use a square area of ​​fixed size, fill it with all 0s, and allow the square area to be outside the image;

[0085] Input the images of the test set into the YOLOV8 target detection model to obtain the predicted box, compare the predicted box position with the real box annotated by the original data, and calculate the loss based on the comparison result;

[0086] While retaining the original classification loss function, a detection box loss function Concentrer-IOU is proposed.

[0087] ① Perform Cutout data enhancement on the training set images. First, randomly read images from the dataset, use a square area of ​​fixed size, fill it with all 0s, and allow the square area to be outside the image.

[0088] ② Input the image of the test set into the target detection model to get the predicted box, compare the position of the predicted box with the real box annotated by the original data, and calculate the loss based on the comparison result

[0089] ③Use Concentrer-IOU as the detection box loss function.

[0090] ④ Combine the two loss functions, perform supervised training on the detection network, adjust the hyperparameters and save the best model file as the test model; finally, save the model file to the processor for detection.

[0091] First calculate the SIOU function. SIOU needs to calculate four loss functions: angle loss ∧, distance loss Δ, shape loss Ω, and IOU loss.

[0092] Basically, the model will try to make predictions first on the X or Y axis (whichever is closest), and then continue along the relevant axis. To achieve this, the convergence process will first try to minimize the angle a if a ≤ 4, otherwise minimize

[0093] The output information after the operation includes four categories: object category information, object detection box information, object category confidence, and the distance between the object and itself.

[0094] Furthermore, the loss calculation includes the following steps:

[0095] Calculate the angle loss ∧,

[0096]

[0097] Where N is the distance between detection boxes, c h is the height of the rectangle with L as the diagonal; a is the angle between the width and the diagonal, and b is the angle between the height and the diagonal; is the center coordinate of the real frame, is the center coordinate of the prediction box;

[0098] Calculate the distance loss Δ:

[0099]

[0100] Where x and y are the horizontal and vertical coordinates of the center point of the detection frame respectively; c ω is the width of the rectangle with L as the diagonal; γ is a constant;

[0101] Calculate the shape loss Ω:

[0102]

[0103] Where ω is the width of the prediction box, ω gt is the width of the true box, h is the height of the predicted box, h gt is the height of the real box; θ is a constant;

[0104] The value of θ is a very important term in this equation, which controls how much attention is needed for the shape loss. The definition range of θ is 2 to 6. If the value of θ is set to 1, the shape will be optimized immediately, thus affecting the free movement of the shape. In this embodiment, the value of θ is defined as 3 or 4.

[0105] Finally, we calculate L SIoU :

[0106] Calculate L Concentrer―IoU :L Concentrer―IoU =1-IoU Concentrer ;

[0107]

[0108] d and u are constants. In this embodiment, d is 0.00 and u is 0.95.

[0109] Among them, A is the intersection of the real box and the predicted box, and B is the union of the real box and the predicted box;

[0110] L Concentrer―SIoU =L SIoU +IoU―IOU Concentret ;

[0111] L 检测框 =L Concentrer―SIoU ;

[0112] L=L 检测框 +L 分类损失 .

[0113] Furthermore, preprocessing the road condition image includes the following steps:

[0114] Scaling the road condition image, changing the channel order of the road condition image, converting the road condition image into a CHW format, and then expanding the dimension of the road condition image and converting it into a four-dimensional array BCHW;

[0115] The storage mode of the traffic condition images in the memory is changed to continuous storage, and the precision of the traffic condition images is converted to single precision;

[0116] The road condition image is normalized and input into the obstacle recognition model. The output information of the obstacle recognition model is subjected to non-maximum suppression processing, and the obstacle coordinates and the distance between the obstacle and itself are output.

[0117] Furthermore, the obstacle avoidance instruction includes:

[0118] Calculate the priority of each obstacle based on the object category information of the obstacle and the distance between the obstacle and itself;

[0119] Based on the priorities, obstacles are avoided in order.

[0120] If there is an obstacle less than 3 meters away from itself, the processor will send the type, direction and distance of the obstacle to the central processor, which will then perform obstacle avoidance.

[0121] For obstacles that are close and have a higher priority, the obstacle avoidance module will change the direction or speed of the wheelchair through the motor control system to bypass the obstacle. It will stop and avoid dynamic obstacles, while it will actively bypass static obstacles.

[0122] For the recognition of dynamic obstacles, the model can use continuous image frames to detect the movement of obstacles. If the position of an obstacle changes significantly in continuous image frames, the model can identify it as a dynamic obstacle.

[0123] The following example illustrates obstacle avoidance instructions: the system assigns a basic priority score to each obstacle (different obstacle object categories correspond to different basic priority scores), which is based on the potential danger that the obstacle may cause to the user. For example:

[0124] Pedestrians: basic priority score is 60;

[0125] Bicycle: Basic priority score is 70;

[0126] Automobile: The basic priority score is 90;

[0127] Trees: Base priority score is 50;

[0128] Telephone poles: base priority score is 60;

[0129] Next, the system considers the distance of the obstacle from the user. The closer the distance, the higher the priority score. For example, the system may use the following rules:

[0130] Distance < 1 meter: distance score is 100;

[0131] 1 meter ≤ distance < 2 meters: distance score is 80;

[0132] 2 meters ≤ distance < 3 meters: distance score is 60;

[0133] Distance ≥ 3 meters: distance score is 40;

[0134] Finally, the priority of an obstacle can be calculated by adding the base priority score to the distance score. For example, if a bicycle is 1.5 meters away from the user, its priority is calculated as follows:

[0135] Bicycle Basic Priority Score: 80;

[0136] Distance score (1 meter ≤ distance < 2 meters): 80;

[0137] Total priority score: 80+80=160;

[0138] The system calculates the priority of each obstacle based on its type and distance, and avoids it accordingly. The higher the total priority score, the higher the priority it will be avoided.

[0139] Example 2

[0140] This embodiment provides an intelligent blind-guiding balancing wheelchair, which is used to implement an intelligent blind-guiding method as described in Embodiment 1, including:

[0141] The wheelchair body is used to move on the travel route;

[0142] A voice module, used to receive voice input and send voice feedback instructions;

[0143] A route planning module is used to generate N travel routes based on the initial location point and the target point;

[0144] The visual module is used to obtain real-time road condition images during navigation;

[0145] An obstacle avoidance module, used for determining whether there is an obstacle in the road condition based on the road condition image;

[0146] A balance sensing module is used to sense whether the wheelchair is in a balanced state;

[0147] A motor control module is used to keep the wheelchair body in a balanced state;

[0148] The central processing module is used to connect, manage and control the voice module, route planning module, vision module, obstacle avoidance module, balance perception module and motor control module.

[0149] Specifically, the intelligent guiding balancing wheelchair includes a wheelchair body, a central processor installed on the wheelchair body, a positioning system, a route planning system, an obstacle avoidance system, a balance sensing system, a motor control system, a BMS power supply system, and a voice system.

[0150] like Figure 1As shown, the arrows in the figure indicate the direction of data flow. The central processor of the center is responsible for managing and controlling the interaction of the above systems. The route planning system is connected to the central processor via a serial port, the obstacle avoidance system is connected to the central processor via Ethernet, the positioning system is connected to the central processor via a serial port, the balance perception system is connected to the central processor via a serial port, the language system is connected via a serial port, the BMS power system is connected to the central processor via CAN communication, and the motor control system is connected to the central processor via CAN communication.

[0151] In this embodiment, the positioning system communicates with the central processor using the JSON data format. When communicating with the central processor, the positioning system will send the geographic information read by the positioning system, including geographic location, administrative area query, traffic events, longitude and latitude positioning, IP positioning, and static maps. The positioning system sends coordinate conversion, weather query, and trajectory correction to the central processor. The positioning system obtains geographic information every 100ms and sends the geographic information to the central processor immediately after obtaining the geographic information.

[0152] In this embodiment, the language system communicates with the central processor using the JSON data format. The language system includes a speaker and a microphone. The language system uses a microphone to receive what the user says. The language system uses the Whisper speech recognition model to convert the user's words into text, and uses the Cerebras model to extract the destination keywords in the obtained text and send them to the central processor. When talking to the user, the XFS5152CE speech synthesis module is used to convert the text into a voice signal and output it through the speaker. Before departure, it will ask whether the destination is correct, and after arriving, it will prompt that the destination has been reached.

[0153] In this embodiment, the BMS power supply is composed of 12 groups of single 3V batteries connected in series, with a total voltage of 36V and a maximum current of 30A. The BMS power supply supplies power to all the systems.

[0154] In this embodiment, the route planning system communicates with the central processing unit using the JSON data format. The route planning system transmits network information with Baidu Map by calling the API of Baidu Map. The route planning system will obtain geographic information from the central processing unit and provide it to Baidu Map, allowing Baidu Map to output route parameters such as route overview (total distance, estimated time, etc.), detailed route points (coordinate points, road names, directions, turn instructions, etc.), estimated time (total time and estimated time for each section), real-time traffic information (congestion, accident information, etc.), map images or URLs, errors and status codes, etc. After obtaining all parameters, the time, distance cost, number of transfers, and passenger flow of each route are listed locally, and the route optimization parameters specified by the user are obtained through the central processing unit for route selection.

[0155] In this embodiment, four cameras (i.e., visual modules) are provided at the front, rear, left, and right of the wheelchair. The cameras are used to transmit the captured images to the central processor in real time through the switch network port, and transmit them to the obstacle avoidance module. The obstacle avoidance module obtains obstacle category information and the distance between the obstacle and itself, and calculates the priority of each obstacle. The priority is obtained by the obstacle category information and the distance between the obstacle and itself. For obstacles that are closer and have higher priorities, the obstacle avoidance module will change the direction or speed of the wheelchair through the motor control system to bypass the obstacles. The stop avoidance method is adopted for dynamic obstacles, while the active bypass method is adopted for static obstacles.

[0156] In this embodiment, a gyroscope balance sensor and a motor are provided under the wheelchair seat. When the wheelchair travels on an unstable road surface, the gyroscope balance sensor generates and sends instructions to the motor under the seat, and the motor drives the seat to keep it in a stable state at all times.

[0157] Example 3

[0158] In this embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, an intelligent blind guiding method as described in Embodiment 1 is implemented.

[0159] Example 4

[0160] This embodiment provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements an intelligent blind guiding method as described in Example 1.

Claims

1. An intelligent blind-guiding method, characterized in that: The following steps are involved: Receive a user's voice input, and parse out key geographical location information based on the voice input; send a voice feedback instruction based on the key geographical location information; the voice feedback instruction is used to interact with the user to confirm whether the key geographical location information is information about the destination; if so, perform the following steps: Acquire the current location information and set it as the initial location point, set the geographical location key information as the target point; generate N travel routes based on the initial location point and the target point; Selecting an optimal route from N travel routes, and sending a navigation travel instruction based on the optimal route; During the navigation process, the road condition image is obtained in real time, and whether there is an obstacle in the road condition is determined based on the road condition image: if so, an obstacle avoidance instruction is sent; otherwise, a maintain status quo instruction is sent, and the maintain status quo instruction is to continue moving and continuously obtain and analyze the road condition image until the target point is reached.

2. An intelligent blind guiding method as claimed in claim 1, characterized in that: Selecting the best route from N travel routes includes the following steps: Obtaining route optimization parameters for each travel route; the route optimization parameters at least include: route time, travel expenses, number of transfers, and passenger flow; The route preference parameters are scored and assigned corresponding weight parameters; based on the scores and weight parameters, a preference score for each travel route is calculated; and based on the preference scores, an optimal route is determined from the N travel routes.

3. An intelligent blind guiding method as claimed in claim 1, characterized in that: Judging whether there is an obstacle in the road condition based on the road condition image includes the following steps: The road condition image is preprocessed, the preprocessed road condition image is input into an obstacle recognition model, and obstacle information is output; the obstacle information at least includes object category information, object detection frame information, object category confidence, and the distance between the object and itself.

4. An intelligent blind guiding method as claimed in claim 3, characterized in that: The construction process of the obstacle recognition model is as follows: Obtaining obstacle images existing on the road, and annotating the obstacle images; dividing the annotated obstacle images into training set images and test set images; enhancing the training set images, and inputting the test set images into a target detection model to obtain a prediction frame; A comparison is performed based on the position of the predicted frame and the position of the real frame annotated in the obstacle image, and a loss calculation is performed according to the comparison result; based on the loss calculation result, a target detection model is trained to obtain the obstacle recognition model.

5. An intelligent blind guiding method as claimed in claim 4, characterized in that: The loss calculation includes the following steps: Calculate the angle loss ∧, Where N is the distance between detection boxes, c h is the height of the rectangle with L as the diagonal; a is the angle between the width and the diagonal, and b is the angle between the height and the diagonal; is the center coordinate of the real frame, is the center coordinate of the prediction box; Calculate the distance loss Δ: Where x and y are the horizontal and vertical coordinates of the center point of the detection frame respectively; c ω is the width of the rectangle with L as the diagonal; γ is a constant; Calculate the shape loss Ω: Where ω is the width of the prediction box, h is the width of the real box, and h is the height of the prediction box. gt is the height of the real box; θ is a constant; Finally, we calculate L SIoU : Calculate L Concentrer―IoU :L Concentrer―IoU =1-IoU Concentrer ; Among them, A is the intersection of the real box and the predicted box, and B is the union of the real box and the predicted box; L Concentrer―SIoU =L SIoU +IoU―IoU Concentrer ; L 检测框 =L Concentrer―SIoU ; L=L 检测框 +L 分类损失 .

6. An intelligent blind guiding method as claimed in claim 3, characterized in that: Preprocessing the road condition image includes the following steps: Scaling the road condition image, changing the channel order of the road condition image, converting the road condition image into a CHW format, and then expanding the dimension of the road condition image and converting it into a four-dimensional array BCHW; The storage mode of the traffic condition images in the memory is changed to continuous storage, and the precision of the traffic condition images is converted to single precision; The road condition image is normalized and input into the obstacle recognition model. The output information of the obstacle recognition model is subjected to non-maximum suppression processing, and the obstacle coordinates and the distance between the obstacle and itself are output.

7. An intelligent blind guiding method as claimed in claim 3, characterized in that: The obstacle avoidance instruction includes the following process: Calculate the priority of each obstacle based on the object category information of the obstacle and the distance between the obstacle and itself; Based on the priorities, obstacles are avoided in order.

8. An intelligent guiding wheelchair, characterized in that: A method for implementing an intelligent blind guiding method as claimed in any one of claims 1 to 7, comprising: The wheelchair body is used to move on the travel route; The voice module is used to receive the user's voice input and send voice feedback instructions; A route planning module is used to generate N travel routes based on the initial location point and the target point; The visual module is used to obtain real-time road condition images during navigation; An obstacle avoidance module, used for determining whether there is an obstacle in the road condition based on the road condition image; A balance sensing module is used to sense whether the wheelchair is in a balanced state; A motor control module is used to keep the wheelchair body in a balanced state; The central processing module is used to connect, manage and control the voice module, route planning module, vision module, obstacle avoidance module, balance perception module and motor control module.

9. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an intelligent blind guiding method as claimed in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, it implements an intelligent blind guiding method as described in any one of claims 1 to 7.

Images