Path planning method for tennis ball picking robot and tennis ball picking robot

Abstract

The invention relates to a path planning method for a tennis ball picking robot and the tennis ball picking robot. The robot carries out background modeling, parameter setting and model loading in sequence; image acquisition and background modeling updating are carried out firstly, then tennis ball target preliminary extraction is carried out, and finally tennis ball target confirmation is carried out, namely whether a tennis ball is detected or not; if the tennis balls are detected, the robot moves nearby and completes ball picking until all the tennis balls are picked; if no tennis ball is detected, adding one to the value of the counter; if the value of the counter is larger than the first preset value, the robot rotates clockwise by a second preset value; if the value of the counter is not larger than the first preset value, whether the tennis ball exists or not continues to be detected; if the rotation angle of the robot is greater than a third preset value, stopping working; if the rotation angle of the robot is not larger than the third preset value, whether the tennis ball exists or not continues to be detected. Compared with a traditional A * algorithm and a dynamic window algorithm, the method is higher in decision making speed and lower in memory occupation.

Description

Technical Field

[0001] The present invention relates to the technical field of a tennis ball picking robot, and in particular to a path planning method for a tennis ball picking robot and the tennis ball picking robot. Background Art

[0002] Tennis is an elegant and exciting sport, which is widely loved around the world. More and more young people are participating in the sport of tennis and showing great enthusiasm for it. However, there is a problem of time-consuming and laborious ball picking in the tennis training process. At present, there is no effective fully automatic tennis ball picking robot in the domestic market.

[0003] The invention patent with publication number CN116810790A discloses a tennis ball picking service robot based on deep learning, including a host computer, a control mainboard, a binocular camera, a robot chassis and a tennis ball picking mechanism, wherein the tennis ball picking mechanism is plugged into the robot chassis, and the host computer is based on the Yolov5s recognition algorithm and a stereo matching algorithm of polar line constraints. Through the stereo matching algorithm based on the Yolov5s recognition algorithm and the polar line constraints, the three-dimensional coordinates of the target are further calculated based on the binocular parallax principle, thereby improving the tennis ball picking service robot's higher stability in tennis ball recognition, and its accuracy and speed performance are also very excellent. Through a ball picking path planning scheme based on a multi-target A* ball picking path planning algorithm, a more efficient and reasonable ball picking path is planned for the robot. The algorithm takes into account the actual obstacles on the court, so that the planned path is more reasonable, ensuring that the robot will not collide with obstacles during the ball picking stage; the patent has a long path planning time and a high memory usage.

[0004] Therefore, improving a path planning method with short path planning time and low memory usage is an urgent problem to be solved. Summary of the invention

[0005] The purpose of the present invention is to provide a path planning method for a tennis ball picking robot and a tennis ball picking robot in order to overcome the defects of the above-mentioned prior art.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] According to one aspect of the present invention, a path planning method for a tennis ball picking robot is provided, the method comprising the following process:

[0008] S1, the robot enters the initialization phase, and performs background modeling, parameter setting and model loading in sequence;

[0009] S2, after completing the initialization, the robot enters the real-time monitoring stage, first performing image acquisition and background modeling update, then performing preliminary tennis target extraction, and finally performing tennis target confirmation, that is, whether the tennis ball is detected;

[0010] S3, if a tennis ball is detected, the robot moves to the vicinity and completes the picking up of the ball until all tennis balls are picked up;

[0011] S4, if no tennis ball is detected, the value of the counter is increased by one;

[0012] S5. If the value of the counter is greater than the first preset value, the robot rotates clockwise by a second preset value; if the value of the counter is not greater than the first preset value, the robot continues to detect whether there is a tennis ball;

[0013] S6. If the rotation angle of the robot is greater than the third preset value, the robot stops working; if the rotation angle of the robot is not greater than the third preset value, the robot continues to detect whether there is a tennis ball.

[0014] As a preferred technical solution, the parameter settings include settings for median filtering and adaptive threshold segmentation; wherein the operation kernel size of the median filtering is set to 5×5; the neighborhood size of the adaptive threshold segmentation is set to 11, and the constant is set to 2.

[0015] As a preferred technical solution, the first preset value is 30 degrees, the second preset value is 15 degrees, and the third preset value is 720 degrees.

[0016] As a preferred technical solution, the background modeling is established using a Gaussian mixture model.

[0017] As a preferred technical solution, an online update algorithm is used to update the mean of the Gaussian mixture model. The mean is updated using a weighted average method, and the formula is as follows:

[0018] μ new =(1-α)·μ old +α·μ newfit

[0019] where μ old is the original mean, μ newfit is the mean of the new fit, and α represents the weight for controlling the new data.

[0020] According to another aspect of the present invention, there is provided a tennis ball picking robot, comprising a body, a ramp, a robot body, wheels, a ball sweeping plate, a motor, a drawer, a tennis ball recognition unit and a motion control unit, wherein the body, wheels and motion control unit are all mounted on the robot body, the drawer and the tennis ball recognition unit are both mounted on the body, the ramp is connected to the body, the ball sweeping plate and the motor are both mounted on the ramp, the ball sweeping plate is connected to the motor, the tennis ball recognition unit is communicatively connected to the motion control unit, the tennis ball recognition unit captures images and transmits them to the motion control unit, and the motion control unit adopts any of the path planning methods described above.

[0021] As a preferred technical solution, the wheel is a Mecanum wheel.

[0022] As a preferred technical solution, the cross-section of the ball sweeping plate is cross-shaped, and the ball sweeping plate is installed at the entrance of the slope and 1.5 tennis diameters away from the ground; the motor is a DC reduction motor.

[0023] As a preferred technical solution, the robot further includes a handle, the handle is mounted on a drawer, and the drawer is mounted in the vehicle shell.

[0024] As a preferred technical solution, the length and width of the vehicle shell are both greater than the length and width of the robot body.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1. The present invention proposes an efficient path planning method, that is, if the tennis ball target is still not found after waiting for 30 clock cycles, the robot will rotate 15 degrees clockwise; when the rotation angle exceeds 720 degrees (2 circles), it indicates that all the tennis balls in the venue have been picked up, and the robot will stop working. Compared with the traditional A* algorithm and dynamic window algorithm, the decision speed is faster and the memory usage is lower. According to statistics, the average execution time of path planning is reduced by 39% and 19% respectively, and the memory usage is reduced by 65% ​​and 33% respectively.

[0027] 2. The mean update strategy of the Gaussian mixture model of the present invention adopts a weighted average update method, so that the model can gradually adapt to new lighting conditions without excessive adjustment due to sudden changes.

[0028] 3. The DC reduction motor of the present invention ensures motor synchronization through speed feedback control, thereby avoiding the torsional vibration or torsional step-out problem of the sweeping plate due to the asynchronism of the driving motors at the left and right ends.

[0029] 4. In order to facilitate operators to take out balls, the present invention adopts a pull-out method to take out tennis balls, and a handle is provided on the drawer to facilitate pulling out the drawer.

[0030] 5. The cross section of the ball sweeping board of the present invention is cross-shaped, and it rotates under the action of the motor, so that the tennis ball is hit into the slope and enters the drawer. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is the overall workflow diagram of the present invention;

[0032] Figure 2 This is a flow chart of model training of the present invention;

[0033] Figure 3 It is the initialization flow chart of the present invention;

[0034] Figure 4 A tennis ball recognition workflow diagram of the present invention;

[0035] Figure 5 It is a schematic diagram of the overall structure of the present invention;

[0036] Figure 6 The figure is a schematic diagram of the installation of the ball sweeping plate and the motor of the present invention;

[0037] Figure 7 It is a structural schematic diagram of the drawer of the present invention.

[0038] 1. Car shell; 2. Ramp; 3. Robot body; 4. Wheels; 5. Ball sweeper; 6. Motor; 7. Drawer. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0040] To solve the above technical problems, the present invention provides a path planning method for a tennis picking robot and the tennis picking robot; the present invention proposes an efficient path planning method, that is, if the tennis target is not found after waiting for 30 clock cycles, the trolley will rotate clockwise by 15 degrees; when the rotation angle exceeds 720 degrees (2 circles), it indicates that the tennis balls on the site have been picked up, and the robot will stop working. Compared with the traditional A* algorithm and dynamic window algorithm, the decision-making speed is faster and the memory occupancy is lower. After statistics, the average time for one execution of path planning is reduced by 39% and 19% respectively, and the memory occupancy is reduced by 65% and 33% respectively. The mean update strategy of the Gaussian mixture model of the present invention adopts a weighted average method for updating, so that the model can gradually adapt to the new lighting conditions without over-adjusting due to sudden changes. The DC reduction motor of the present invention ensures motor synchronization through a speed feedback control method, avoiding the problem of torsional vibration or torsional out-of-step of the sweeping ball plate due to asynchronous driving motors at both ends. In order to facilitate the operator to pick up the balls, the present invention adopts a pull-out type to take out the tennis balls, and a handle is provided on the drawer to facilitate the drawer to be pulled out. The cross-section of the sweeping ball plate of the present invention is cross-shaped and rotates under the action of the motor to knock the tennis balls into the ramp and make them enter the drawer.

[0041] Embodiment 1

[0042] As Figure 1-4 shown, a path planning method for a tennis picking robot, the method includes the following processes:

[0043] S1. The robot enters the initialization stage and sequentially performs background modeling, parameter setting, and model loading;

[0044] S2. After completing the initialization, the robot enters the real-time monitoring stage. First, it performs image acquisition and background modeling update, then performs preliminary extraction of the tennis target, and finally confirms whether the tennis target is detected, that is, whether a tennis ball is detected;

[0045] S3. If a tennis ball is detected, the robot moves to the vicinity and completes the ball picking until all the tennis balls are picked up;

[0046] S4. If no tennis ball is detected, the value of the counter is incremented by one;

[0047] S5. If the value of the counter is greater than the first preset value, the robot rotates clockwise by the second preset value; if the value of the counter is not greater than the first preset value, continue to detect whether there is a tennis ball;

[0048] S6. If the rotation angle of the robot is greater than the third preset value, stop working; if the rotation angle of the robot is not greater than the third preset value, continue to detect whether there is a tennis ball.

[0049] The parameter settings include settings for median filtering and adaptive threshold segmentation; wherein the operation kernel size of the median filtering is set to 5×5; the neighborhood size of the adaptive threshold segmentation is set to 11, and the constant is set to 2.

[0050] The first preset value is 30 degrees, the second preset value is 15 degrees, and the third preset value is 720 degrees.

[0051] The background modeling is established using a Gaussian mixture model.

[0052] The mean of the Gaussian mixture model is updated using an online update algorithm. The mean is updated using a weighted average method. The formula is as follows:

[0053] μ new =(1-α)·μ old +α·μ newfit

[0054] where μ old is the original mean, μ newfit is the mean of the new fit, and α represents the weight for controlling the new data.

[0055] In this embodiment, the specific implementation method of tennis target recognition is as follows:

[0056] 1. Data collection and annotation stage: Collect a large number of tennis images in different lighting scenarios, covering various lighting conditions indoors and outdoors. Use professional annotation tools to accurately annotate the location, category, and lighting conditions (such as strong light, weak light, etc.) of the tennis balls.

[0057] 2. Model selection and training phase: ResNet-50 is selected as the basic model. According to the characteristics of the tennis ball recognition task, an attention mechanism module is added to the front end of the model to enhance the focus on the key features of the tennis ball. The learning rate is set to 0.001, the Adam optimizer is used, and the training is iterated for 100 epochs. During the training process, the early stopping method is used to prevent overfitting.

[0058] 3. Model testing and optimization stage: In addition to using data enhancement techniques such as random flipping, rotation, and cropping, a generative adversarial network (GAN) is introduced to generate virtual tennis images under different lighting conditions to expand the training data. At the same time, K-fold cross-validation (K=5) is used to evaluate model performance, and the accuracy and generalization ability of the model are improved by adjusting the model hyperparameters. A large number of tests are conducted in tennis ball picking scenes under different lighting conditions, and indicators such as recognition accuracy, false alarm rate, and processing speed are recorded. The algorithm parameters and model structure are adjusted according to the results to ensure accurate and fast recognition of tennis balls.

[0059] 4. Initialization phase: Before the robot starts working, first use the camera to collect multiple sets of court images without tennis. Use the Gaussian Mixture Model (GMM) to process these images, determine multiple Gaussian distributions for each pixel, and establish an initial background model. Set the time interval of the online update algorithm for subsequent updates of the Gaussian mixture model parameters (generally speaking, under relatively stable lighting conditions indoors, the time interval can be set to a few minutes, such as 3-5 minutes. In outdoor environments, since the lighting may change rapidly, especially at sunrise, sunset, or when clouds block the sun, the time interval can be set to 1-2 minutes or even shorter); at the same time, set the relevant parameters such as median filtering and adaptive threshold segmentation. Load the initial background model and the improved convolutional neural network model based on ResNet (Residual Network) trained in the previous steps into the NVIDIA Jetson embedded computing platform. The specific process of loading the convolutional neural network model is: use Python and OpenCV libraries to integrate various modules, deploy models and algorithms on the embedded computing platform, and use CUDA acceleration to improve the inference speed.

[0060] 5. Real-time detection stage: When the robot starts working, it uses the camera to obtain images of the court in real time and transmits the images to the NVIDIA Jetson embedded computing platform. At set time intervals (for example, set to 3 minutes), the parameters (mean, variance) of the Gaussian mixture model are updated using an online update algorithm based on the newly acquired images to adapt to the gradual change of illumination. The specific strategy for updating the parameters of the Gaussian mixture model is: the mean is updated using a weighted average method, and this process can be expressed as μ new =(1-α)·μ old +α·μ newfit , where μ old is the original mean, μ newfitis the mean of the new fit, and α represents the weight for controlling the new data. The variance is also updated in a weighted average manner, so that the model can gradually adapt to the new lighting conditions without over-adjusting due to sudden changes. Then, the real-time camera image is differentially calculated with the current background model to obtain a differential image. The differential image is first subjected to median filtering to remove salt and pepper noise, and then adaptive threshold segmentation is performed to remove small interference areas and preliminarily extract the tennis target. The median filtering and adaptive threshold segmentation (using Gaussian weighting method) are implemented by the medianBlur function and adaptiveThreshold function of the OpenCV library respectively. For the initially extracted tennis target area, normalization and histogram equalization preprocessing are performed to enhance the image contrast. The preprocessed image is input into the convolutional neural network model based on the improved ResNet. The model automatically extracts the features of the tennis ball through the convolutional layer, pooling layer and fully connected layer, and determines whether the area is a tennis ball through the Softmax classifier. If the model determines that the area is a tennis ball, the path planning and picking process is entered; if it is not a tennis ball, it continues to return to the image acquisition step and repeats the operation of the real-time detection stage.

[0061] The specific implementation of robot path planning is as follows:

[0062] When the motion control unit receives the horizontal deviation distance and regional pixel size of the tennis target area through the serial port, it will first move horizontally to make the vehicle body face the tennis target (that is, the tennis target is located on the extension line of the central axis of the car), and then move forward to the side of the tennis ball until the latest read regional pixel size is greater than 550 pixels (after multiple sets of experimental tests, the corresponding tennis target is between 10-15cm from the vehicle body at this time), and then control the ball picking mechanism to start working. The vehicle body will remain motionless until the ball picking mechanism stops working, and will not process the data sent from the serial port. When the ultrasonic sensor installed at the entrance of the tennis storage device cannot detect the new tennis ball entering the tennis storage device, it will control the ball picking mechanism to stop rotating, and the motion control unit will continue to process the data sent from the serial port. If the tennis target is found, the above operation will be repeated and the ball picking path will be replanned. Otherwise, if the tennis target is still not found after waiting for 30 clock cycles, the car will rotate 15 degrees clockwise. When the selected rotation angle exceeds 720 degrees (2 turns), it means that the tennis balls in the venue have been picked up, and the robot will stop working.

[0063] Example 2

[0064] like Figure 5-7As shown, a tennis ball picking robot includes a body 1, a ramp 2, a robot body 3, wheels 4, a ball sweeping plate 5, a motor 6, a drawer 7, a tennis ball recognition unit and a motion control unit, wherein the body 1, wheels 4 and the motion control unit are all mounted on the robot body 3, the drawer 7 and the tennis ball recognition unit are both mounted on the body 1, the ramp 2 is connected to the body 1, the ball sweeping plate 5 and the motor 6 are both mounted on the ramp 2, the ball sweeping plate 5 is connected to the motor 6, the tennis ball recognition unit is communicatively connected to the motion control unit, the tennis ball recognition unit captures images and transmits them to the motion control unit, and the motion control unit adopts any of the path planning methods described above.

[0065] The wheel 4 is a Mecanum wheel. The cross section of the ball sweeper 5 is a cross shape, and the ball sweeper 5 is installed at the entrance of the ramp 2 and 1.5 tennis diameters away from the ground; the motor 6 is a DC reduction motor 6. The robot also includes a handle, which is installed on a drawer 7, and the drawer 7 is installed in the car shell 1. The length and width of the car shell 1 are both greater than the length and width of the robot body 3.

[0066] In this embodiment, the motion control unit adopts the STM32F407VET6 core development board, the image processing unit adopts the NVIDIA Jetson embedded computing platform, with built-in NVIDIA's GPU (Graphics Processing Unit, graphics processor), with CUDA (Compute Unified Device Architecture, a parallel computing platform and programming model) core, supporting a variety of mainstream deep learning frameworks, the car shell 1 and the slope 2 for guiding tennis balls are fixed to the top of the car body by a number of screws, and the tennis ball recognition unit adopts Sony Exmor RS series camera, with high resolution, wide dynamic range, low noise under high sensitivity, fast autofocus, low distortion lens and other performance requirements. The ball sweeping plate 5 is a cross-shaped ball sweeping plate, and a pair of DC reduction motors that control the rotation of the ball sweeping plate are set at the entrance of the slope 2 at a distance of about 1.5 tennis ball diameters from the ground, and a handle is provided at the rear of the drawer 7, which is connected to the rear of the top of the slope 2.

[0067] The present invention is further configured as follows: the motor 6 uses a pair of DC reduction motors to ensure the synchronization of the motor 6 through speed feedback control, thereby avoiding the torsional vibration or torsional desynchronization problem of the sweeping plate 5 due to the asynchronous driving motors 6 at the left and right ends.

[0068] Preferably, in order to facilitate the operator to take out the ball, the tennis ball storage device is connected to the vehicle shell 1 through a sliding guide device, and the tennis ball is taken out in a pull-out manner.

[0069] When the robot body 3 approaches the tennis ball target, the sweeping plate 5 will hit the tennis ball from the bottom of the slope 2 to the top of the slope 2, and then store it in the drawer 7. When the number of tennis balls inside the device reaches a certain amount, the drawer 7 can be pulled out to take out the tennis balls inside. The present invention can significantly reduce the workload of picking up tennis balls, thereby improving the efficiency of tennis training and competition for athletes. The present invention reduces the height and inclination of the slope 2, and improves the success rate of picking up tennis balls; secondly, the sweeping plate 5 is composed of a plurality of nylon plastic rods, which greatly reduces the consumption of raw materials.

[0070] The image processing unit NVIDIA Jetson embedded computing platform and the motion control unit STM32F407VET6 core development board use serial communication to achieve communication. If the image processing unit confirms that the tennis target is detected, the horizontal deviation distance of the tennis target area with the most pixels relative to the center line of the image (the one with the most pixels indicates that it is the tennis target closest to the robot, and the positive or negative horizontal deviation distance represents whether the target area is located in the right front or left front of the car) and the pixel size of the area will be sent to the motion control unit through serial communication, otherwise an empty data packet will be sent.

[0071] The motion control unit is based on the FreeRTOS embedded real-time system to meet the needs of multi-task management and switching during the tennis ball picking process. The tasks that need to be performed during the tennis ball picking process include: system initialization task, image data receiving and parsing task, vehicle body motion control task, ball picking mechanism control task and battery power detection task. The task execution process of each task is as follows:

[0072] (1) System initialization task: After the robot is powered on, this task will power each hardware module, configure the working parameters of each sensor and motor driver, create queues for system communication, semaphores for synchronization between tasks, and flag time groups for controlling access to shared resources. After completing its own workflow, the task will be suspended through the task suspension API function.

[0073] (2) Image data receiving and parsing task: After triggering the DMA overflow interrupt, this task will retrieve the data in the serial port DMA buffer, parse the serial port data packet, and put the data such as the horizontal deviation distance of the tennis target area into the message queue, waiting for subsequent consumption by the vehicle motion control task.

[0074] (3) Vehicle motion control task: This task is mainly responsible for implementing the path planning function after identifying the tennis target. After the task takes out the horizontal deviation distance and regional pixel size of the tennis target area from the message queue, it will first move horizontally to make the vehicle body face the tennis target (that is, the tennis target is located on the extension line of the central axis of the car), and then move forward to the side of the tennis ball until the latest read regional pixel size is greater than 550 pixels (after multiple sets of experimental tests, the corresponding tennis target is between 10-15cm from the vehicle body at this time), and then send a start signal to the ball picking mechanism control task. After receiving the signal, the ball picking mechanism control task will control the ball picking mechanism to start working. The vehicle body will remain motionless until the ball picking mechanism stops working, and the task will not process the messages in the message queue. After receiving the signal to close the ball picking mechanism, the task will continue to process the messages in the message queue. If the tennis target is found, the above operation will be repeated and the ball picking path will be replanned. Otherwise, if the tennis target is still not found after waiting for 30 clock cycles, the car will rotate 15 degrees clockwise. When the rotation angle exceeds 720 degrees (2 circles), it means that all the tennis balls in the venue have been picked up and the robot will stop working.

[0075] (4) Ball picking mechanism control task: After receiving the start signal sent by the vehicle motion control task, this task will control the ball picking mechanism to start working. When the ultrasonic sensor installed at the entrance of the tennis ball storage device cannot detect a new tennis ball, it will control the ball picking mechanism to stop rotating and send a ball picking mechanism shutdown signal to the vehicle motion control task, indicating that path planning can be restarted.

[0076] (5) Battery power detection task: This task will detect the battery power of the robot at regular intervals. When the battery power is low, it will switch to low-power working mode and sound an alarm through a buzzer.

[0077] Taking into account the excellent omnidirectional motion capability of Mecanum wheels, the forward, backward, lateral translation and oblique motion of the vehicle body are innovatively used to replace the traditional steering motion, thereby effectively avoiding the under-turn or over-turn problems that may occur during the steering process of the vehicle.

[0078] The height of the ball sweeping plate 5 from the ground is between 1.5 and 2 tennis diameters, and the height from the slope 2 is between 1 and 1.5 tennis diameters. The length and width of the car shell 1 are slightly larger than the length and width of the car body 3, and the height is between 12 and 14 cm. The tennis recognition unit is a Sony IMX477 camera, which is set in the middle of the top of the car shell 1, and the height from the top of the car shell is between 10 and 20 cm. In order to meet the needs of smoothly hitting the tennis ball into the drawer 7, the maximum speed of the DC reduction motor 6 is 68 rpm, and the maximum torque is 25 kg. The model of the Mecanum wheel encoder motor driver is AT8236, and the model of the DC reduction motor driver is L298N.

[0079] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A path planning method for a tennis ball picking robot, characterized in that: The method comprises the following steps: S1, the robot enters the initialization phase, and performs background modeling, parameter setting and model loading in sequence; S2, after completing the initialization, the robot enters the real-time monitoring stage, first performing image acquisition and background modeling update, then performing preliminary tennis target extraction, and finally performing tennis target confirmation, that is, whether the tennis ball is detected; S3, if a tennis ball is detected, the robot moves to the vicinity and completes the picking up of the ball until all tennis balls are picked up; S4, if no tennis ball is detected, the value of the counter is increased by one; S5. If the value of the counter is greater than the first preset value, the robot rotates clockwise by a second preset value; if the value of the counter is not greater than the first preset value, the robot continues to detect whether there is a tennis ball; S6. If the rotation angle of the robot is greater than the third preset value, the robot stops working; if the rotation angle of the robot is not greater than the third preset value, the robot continues to detect whether there is a tennis ball.

2. A path planning method for a tennis ball picking robot according to claim 1, characterized in that: The parameter settings include settings for median filtering and adaptive threshold segmentation; wherein the operation kernel size of the median filtering is set to 5×5; the neighborhood size of the adaptive threshold segmentation is set to 11, and the constant is set to 2.

3. A path planning method for a tennis ball picking robot according to claim 1, characterized in that: The first preset value is 30 degrees, the second preset value is 15 degrees, and the third preset value is 720 degrees.

4. A path planning method for a tennis ball picking robot according to claim 1, characterized in that: The background modeling is established using a Gaussian mixture model.

5. A path planning method for a tennis ball picking robot according to claim 4, characterized in that: The mean of the Gaussian mixture model is updated using an online update algorithm. The mean is updated using a weighted average method. The formula is as follows: m new =(1-a)·m old +a·m newfit where μ old is the original mean, μ newfit is the mean of the new fit, and α represents the weight for controlling the new data.

6. A tennis ball picking robot, characterized in that: The invention comprises a vehicle shell (1), a ramp (2), a robot body (3), wheels (4), a ball sweeping plate (5), a motor (6), a drawer (7), a tennis ball recognition unit and a motion control unit, wherein the vehicle shell (1), the wheels (4) and the motion control unit are all mounted on the robot body (3), the drawer (7) and the tennis ball recognition unit are all mounted on the vehicle shell (1), the ramp (2) is connected to the vehicle shell (1), the ball sweeping plate (5) and the motor (6) are all mounted on the ramp (2), the ball sweeping plate (5) and the motor (6) are connected, the tennis ball recognition unit is in communication connection with the motion control unit, the tennis ball recognition unit collects images and transmits them to the motion control unit, and the motion control unit adopts the path planning method according to any one of claims 1 to 5.

7. A tennis ball picking robot according to claim 6, characterized in that: The wheel (4) is a Mecanum wheel.

8. The tennis ball picking robot according to claim 6, characterized in that: The cross section of the ball sweeping plate (5) is cross-shaped; the ball sweeping plate (5) is installed at the entrance of the slope (2) and is 1.5 tennis ball diameters away from the ground; and the motor (6) is a DC reduction motor (6).

9. The tennis ball picking robot according to claim 6, characterized in that: The robot further comprises a handle, wherein the handle is mounted on a drawer (7), and the drawer (7) is mounted in the vehicle shell (1).

10. The tennis ball picking robot according to claim 6, characterized in that: The length and width of the vehicle shell (1) are both greater than the length and width of the robot body (3).

Images