Ball picking split robot capable of intelligently recognizing quality of table tennis balls and working method

Abstract

The invention belongs to the technical field of intelligent robots, and discloses a ball picking split robot capable of intelligently recognizing whether table tennis balls are good or not and a working method thereof.The ball picking split robot comprises a base and a ball picking robot, the base is divided into three layers, and the three layers are a ball storage box, a storage box and a charging bin in sequence from top to bottom; the base is provided with a vertically arranged pipeline; the ball picking robot comprises a battery, universal wheels, a shell, a driving motor, a microcomputer, a collecting barrel, a collector, a transition part arranged between the collecting barrel and the shell, two crawler belts arranged in the shell and perpendicular to each other, and a laser radar sensor. A first storage area and a second storage area are further arranged in the shell, and the second storage area is arranged close to the transition part; piston push rods are arranged on one side of the crawler in the width direction; and a fan system is arranged on the same side of the piston push rod. The ball picking split robot solves the problems that in table tennis training, manual table tennis picking efficiency is low, and time and labor are consumed.

Description

Technical Field

[0001] The invention relates to the technical field of intelligent robots, and in particular to a ball-picking split robot capable of intelligently identifying the quality of table tennis balls and a working method thereof. Background Art

[0002] In traditional table tennis training, picking up the ball is a time-consuming and labor-intensive process. Especially in high-intensity training, athletes and coaches need to spend a lot of time picking up the ball, which not only reduces the efficiency of training, but also affects the continuity and quality of training.

[0003] Chinese invention patent: Announcement number is "CN115253227B", and the name is "A table tennis ball automatic tracking and sorting robot", which discloses a table tennis ball automatic tracking and sorting robot, including a robot box, a picking device, a classification device, a mobile component, and a ball storage device; the box shell is provided with an ultrasonic sensor, and the upper part and one side of the box shell are connected with an L-shaped picking device, and the picking device includes a fan and a fan conveying channel; the fan conveying channel includes an L-shaped conveying channel, a ball outlet, a ball inlet, and a sensor group, and the sensor group is an image sensor and a sound sensor, and the bottom of the robot box is connected with a mobile component: the robot box is provided with a classification device, and a pair of ball storage devices are provided at the bottom of the classification device; in view of the low degree of automation of existing table tennis robots and the inability to collect and classify without loss. This technical solution collects table tennis balls without loss through a fan and an L-shaped conveying channel, uses a sensor group to identify the integrity, and automatically classifies and stores them, realizing automatic picking, obstacle avoidance, identification and classification recovery, reducing the safety hazards of balls scattered on the ground, and reducing labor costs. However, this technical solution only classifies and recycles the table tennis balls, and cannot provide athletes with the function of taking the table tennis balls independently. At the same time, this technical solution does not explain how to identify the quality of the table tennis balls, and cannot meet the needs. Summary of the invention

[0004] In order to solve the problem that the sorting robot in the prior art cannot provide athletes with the function of autonomously picking up table tennis balls and does not explain how to identify the quality of table tennis balls, the present invention proposes a ball-picking split robot that can intelligently identify the quality of table tennis balls, aiming to improve the training efficiency and quality of athletes, and also provide athletes and coaches with a more convenient and comfortable training environment.

[0005] The present invention is achieved through the following technical solutions: it includes a base and a ball picking robot, the base is a box structure, the base is divided into three layers, which are a ball storage box, a storage box and a charging bin from top to bottom, the top of the ball storage box is open and a top cover is provided; a vertically arranged pipe that passes through the ball storage box, the storage box and the charging bin and is connected to the ball picking robot is also provided inside the base; the ball picking robot includes a battery arranged at the bottom, a universal wheel arranged at the bottom of the battery, a shell arranged at the top of the battery, a driving motor and a microcomputer; the ball picking robot also includes a collecting bucket arranged on one side of the battery, a collector arranged inside the collecting bucket and rotatably connected to the collecting bucket, a transition part arranged between the collecting bucket and the shell, two tracks arranged inside the shell and arranged perpendicular to each other, a camera arranged at the junction of the two tracks, and a camera arranged at the top of the shell and against the A laser radar sensor is arranged near the transition part; an opening is provided at the end of the collecting barrel away from the battery; a first storage area and a second storage area are also arranged inside the shell, and the second storage area is arranged near the transition part; one of the tracks is arranged near the transition part and parallel to the transition part, and the other track is arranged perpendicular to the transition part and passes through the second storage area and the first storage area in sequence; a piston push rod is arranged on the side away from the second storage area in the width direction of the track arranged perpendicular to the transition part, and the piston push rod is arranged opposite to the second storage area; a fan system is arranged on the same side of the piston push rod opposite to the first storage area, and the fan system is also arranged at the bottom of the first storage area; the microcomputer is electrically connected to the battery, and is connected to the drive motor, the fan system, the piston push rod, the camera and the laser radar sensor through signals; the drive motor is connected to the universal wheel, the collector and the track.

[0006] The present invention also provides a working method of the ball-picking split robot for intelligently identifying the quality of table tennis balls according to the present invention, comprising the following steps:

[0007] S1, the laser radar sensor emits a laser beam, receives the reflected laser signal, and sends the laser signal to a microcomputer;

[0008] S2, the microcomputer processes the laser signal to obtain the optimal path;

[0009] S21, the microcomputer obtains the distance to each point in the environment according to the round-trip time of the laser beam and generates point cloud data;

[0010] S22, preprocessing the point cloud data, performing point cloud registration, and performing environmental modeling again to construct a three-dimensional model of the environment surrounding the ball picking robot;

[0011] S23, detecting and classifying obstacles according to the three-dimensional model obtained in step S22, and distinguishing between table tennis balls and obstacles;

[0012] S24, performing dynamic path planning according to the DynamicA* algorithm;

[0013] S241, initialize and define the starting point S and the target point T, initialize the open list OpenList, add the starting point S to it, set g=0, f=h, where h is calculated according to the heuristic function, and initialize the closed list ClosedList, which is initially empty;

[0014] S242, calculate the evaluation function f(m) of each neighbor node m of node n. If node m is in ClosedList, skip the evaluation function calculation of this node, take the node m corresponding to the minimum value, put it into OpenList, put the node n originally in OpenList into ClosedList, and update the newly put node m to node n. Repeat this step until node n is the target point T. At this time, the path composed of all coordinates in ClosedList is the optimal path P. The calculation formula of the evaluation function f(m) is as follows:

[0015] f=g+h

[0016] g=g+c*d

[0017] Where g and g are the actual costs from the starting point to nodes n and m respectively; h is the heuristic estimated cost from node m to the target point; c is the cost coefficient, if the area where node m is located is an ordinary floor, the cost coefficient is c1, if the area where node m is located is cluttered, the cost coefficient is c2; d is the distance from node n to node m;

[0018] S243, if a new obstacle is detected at node k, update the cost of the affected node, take the node where the ball-picking robot is located as the initial node, repeat step S242, and obtain a new optimal path P2;

[0019] S3, the ball picking robot performs table tennis ball collection according to the optimal path;

[0020] S4, when the collector contacts the ping-pong ball, it starts to rotate, and sends the ping-pong ball into the housing through the transition portion;

[0021] S5, the table tennis ball that enters the housing falls on the crawler arranged parallel to the transition part, is transported by the crawler to the crawler arranged perpendicular to the transition part, and is then transported to the camera by the crawler arranged perpendicular to the transition part. The camera takes an image of the table tennis ball and sends it to the microcomputer;

[0022] S6, the microcomputer distinguishes good and bad balls. If it is a bad ball, it sends a command to the piston push rod, and the piston push rod pushes the bad ball into the second storage area; if it is a good ball, it sends a command to the fan system to send the good ball into the first storage area;

[0023] S61, the microcomputer extracts key features from the image and preprocesses the image to obtain a preprocessed image;

[0024] S62, using the Canny edge detection algorithm to obtain the best edge detection result;

[0025] S621, for each small block of the preprocessed image, calculate the local variance according to the three RGB channels respectively, the formula is as follows:

[0026]

[0027] Among them, r ij is the pixel value of the small block, i and j are integers greater than 0; is the average value; is the variance;

[0028] S622, selecting a corresponding Gaussian kernel size according to the variance obtained in step S621;

[0029] S623, preliminarily detecting the edge area in the image, adjusting the pixels within 3 pixels from the edge to 0.5 times the weight of the original 3x3 Gaussian kernel, and obtaining an image after adaptive Gaussian filtering;

[0030] S624, using the Scharr operator to perform convolution operations in the horizontal and vertical directions on the image after the adaptive Gaussian filtering, respectively, to obtain a horizontal gradient component Gx and a vertical gradient component Gy;

[0031] S625, select two scales, the small scale is the original image size, the large scale is the original image downsampled to half of the original size, use the Scharr operator on the small scale image to calculate the gradient components Gx1, Gy1, and calculate Gx2, Gy2 on the large scale image; fuse the gradient information, using the weighted average method; the gradient amplitude after fusion Calculate the average gray value of the entire ping pong ball image And the standard deviation σG formula is as follows: where g i is the grayscale value of the pixel in the image, N is the total number of pixels in the image,

[0032] S626, set high threshold T h =G+1.5σ G , low threshold T l =G-0.5σ G ; Perform the first edge detection and count the ratio of the number of edge pixels detected to the total number of pixels; if the ratio is less than 5%, adjust the high threshold to T h =T h -5, the low threshold is adjusted to Tl =T l -5; if the ratio is greater than 30%, the high threshold is adjusted to T h =T h +5, low threshold adjusted to T l =T l +5, perform edge detection again to obtain the best edge detection result;

[0033] S63, observing the edge contour obtained by the best edge detection result, if the edge contour is continuous and smooth without obvious breakpoints or jagged shapes, and after fitting the edge contour, the deviation between the radius of the circle obtained by fitting and the radius of the ideal table tennis ball is less than 10%, then it is a good ball, and the microcomputer sends a command to the fan system to send the good ball to the first storage area; otherwise, it is a bad ball, and the microcomputer sends a command to the piston push rod to send the bad ball to the second storage area;

[0034] S7. When the ping-pong balls in the ball-picking robot are full or the battery power is less than 10% of the total power, the ball-picking robot returns to the base to charge; the fan system at the bottom of the first storage area works to blow the good balls in the first storage area into the ball storage box along the pipeline.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. The ball picking split robot designed by the present invention solves the problem of low efficiency, time-consuming and labor-intensive human picking of table tennis balls during table tennis training.

[0037] 2. The ball-picking split robot designed by the present invention is equipped with an intelligent recognition device, including a high-definition camera and an image recognition system. The high-definition camera can capture the state of the table tennis ball in all directions. The image recognition system has a built-in processing unit and a set of standard table tennis ball feature databases. The processing unit will compare the extracted features with the standard features in the database. The database stores standard features and needs to be updated and maintained regularly to adapt to changes in table tennis manufacturing standards or system upgrades.

[0038] 3. The ball-picking split robot designed by the present invention is an integrated intelligent table tennis ball picking device. It integrates multiple functions such as table tennis ball collection, identification, and classified storage. Through sensor technology and intelligent algorithms, the robot can accurately sense the position and state of table tennis balls and achieve efficient collection and sorting.

[0039] 4. The ball-picking split robot designed by the present invention is provided with a collection device, which can accurately determine the position of the table tennis ball according to the information collected by the sensor, and realize flexible and free movement in combination with an efficient driving device. Whether on the court or in other complex environments, the robot can quickly and accurately collect the table tennis balls into the robot, ensuring the timely recovery and sorting of the table tennis balls.

[0040] 5. The ball-picking split robot designed by the present invention has a built-in intelligent recognition module, which can distinguish good balls from bad balls and store them separately. This not only facilitates subsequent processing and use, but also improves the management efficiency and quality of table tennis.

[0041] 6. The ball-picking split robot designed by the present invention has a built-in path planning module and is equipped with a variety of trained algorithm models. These algorithm models enable the robot to effectively realize automatic path planning and automatic obstacle avoidance functions, allowing it to move freely in complex environments without being affected by obstacles. The robot's work efficiency and adaptability are improved, allowing it to operate stably in various scenarios. At the same time, the module can also collect the detailed position of the table tennis ball, plan the optimal ball-picking path, and ensure the maximum efficiency of ball-picking.

[0042] 7. The ball picking split robot designed by the present invention is of split type, including a ball picking robot and a base. The ball picking robot collects and sorts table tennis balls and places good balls in the base for athletes to take independently. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the overall structure of the present invention.

[0044] Figure 2 It is a schematic diagram of the overall structure of the base of the present invention.

[0045] Figure 3 The figure is a schematic diagram of the overall structure of the ball picking robot of the present invention.

[0046] Figure 4 The figure is a schematic diagram of the chassis structure of the ball picking robot of the present invention.

[0047] Figure 5 It is a schematic diagram of the collecting device of the ball picking robot of the present invention.

[0048] Indicated in the figure:

[0049] 1. Collection bucket; 2. Collector; 3. Transition part; 4. Track; 5. Piston push rod; 6. Fan system; 7. LiDAR sensor; 8. Microcomputer; 9. Camera; 10. First storage area; 11. Second storage area; 12. Ball storage box; 13. Universal wheel; 14. Drive motor; 15. Battery; 16. Magnetic charger; 17. Baffle; 18. Pipeline. DETAILED DESCRIPTION

[0050] Advantages and features of the present invention will be illustrated and explained by the following non-limiting description of preferred embodiments thereof, which are given by way of example only with reference to the accompanying drawings.

[0051] like Figures 1 to 5As shown, the present invention provides a split ball-picking robot that can intelligently identify good and bad table tennis balls, which can automatically collect and classify table tennis balls on the ground during table tennis training, thereby improving the efficiency of training. The split ball-picking robot includes a base and a ball-picking robot. The base is a box structure, and the base is divided into three layers, which are a ball storage box 12, a storage box, and a charging compartment from top to bottom. The top of the ball storage box 12 is open and provided with a top cover, which is used to store good balls collected by the ball-picking robot; a double door is provided on the side of the storage box, which is used to store the athlete's clothes, etc.; a magnetic charger 16 is provided at the charging compartment, which is used to dock with the ball-picking robot for charging. A pipe 18 is also provided inside the base. The pipe 18 is vertically provided and passes through the ball storage box 12, the storage box and the charging bin and is connected to the ball picking robot. The pipe 18 is used to guide the good balls sorted by the ball picking robot into the ball storage box 12. In order to make the good balls fall into the ball storage box 12 more smoothly, a baffle 17 is provided on the top of the pipe 18 to reduce the impact force of the table tennis balls and ensure their integrity and safety. The baffle 17 is preferably in an arc shape with the center of the circle facing the pipe 18.

[0052] The ball picking robot includes a battery 15, which is arranged at the bottom of the ball picking robot as the base of the ball picking robot to keep the center of gravity of the robot stable. At the same time, the battery 15 is the main energy supply device of the ball picking robot, has a high energy density and a long service life, and can provide continuous and stable power for the driving motor 14, intelligent control system and other electronic devices of the ball picking robot. The battery 15 cooperates with the magnetic charger 16 to complete the charging of the ball picking robot. When the power of the ball picking robot is about to run out, the ball picking robot will automatically return to the base for charging. A shell is arranged on the top of the battery 15, and the shell wraps the remaining parts of the ball picking robot for protection. A microcomputer 8 is also arranged on the top of the battery 15. The microcomputer 8 is electrically connected to the battery 15, and is connected to the driving motor 14, the fan system 6, the piston push rod 5, the camera 9 and the laser radar sensor 7 through signals, and is used to receive and send various signals to control the action of the ball picking robot.

[0053] The bottom of the battery 15 is provided with a universal wheel 13, which enables the ball picking robot to move freely in any direction. The universal wheel 13 is connected to a drive motor 14 arranged on the top of the battery 15, and is powered by the drive motor 14. In the ball picking robot, the universal wheel 13 enables the ball picking robot to move flexibly in the table tennis field and quickly and accurately reach the position of the ball. The drive motor 14 adjusts the speed and direction as needed. The drive motor 14 is connected to the universal wheel 13 through a reducer and a transmission system to provide sufficient torque and precise control. When the ball picking robot needs to move, the microcomputer 8 will send an instruction to the drive motor 14. The drive motor 14 adjusts the speed and direction according to the instruction, thereby driving the universal wheel 13 to rotate.

[0054] like Figure 3 and Figure 5 As shown, the ball picking robot also includes a collection bucket 1 arranged on one side of the battery 15, and a transition portion 3 is arranged between the collection bucket 1 and the outer shell of the ball picking robot. The collection bucket 1 is provided with an opening at one end away from the battery 15 for the table tennis ball to pass through. The collection bucket 1 is provided with a collector 2 for rotation inside, and the collector 2 can transport the table tennis ball in the collection bucket 1 to the transition portion 3 by rotating, and then finally enter the inner shell of the ball picking robot through the guiding effect of the transition portion 3. Preferably, the collector 2 is composed of a plurality of groups of blades arranged in an axial array along the collection bucket 1, and each group of blades is composed of three blades arranged in a circumferential array. Each blade has a certain curvature, which can fit the curvature of the table tennis ball, thereby protecting the table tennis ball from deformation caused by mechanical collision and other problems. At the same time, the structure of the collector 2 is simple, making its maintenance and cleaning work convenient, greatly reducing the maintenance cost and time. The collector 2 is made of high-quality materials, which not only ensures its durability under long-term and high-intensity use, but also avoids damage to the collector 2 caused by material fatigue, thereby extending the service life of the equipment. The transition part 3 is wide at the part connected to the collection bucket 1, narrow at the part connected to the shell, and the transition part 3 is tilted so that the part connected to the collection bucket 1 is lower than the part connected to the shell, which can effectively avoid the accumulation of a large number of table tennis balls during the collection process, thereby reducing the mechanical failure that may be caused by excessive accumulation, and the table tennis balls can slide quickly and orderly into the shell, ensuring the smooth and efficient collection process of the table tennis balls. The collector 2 is connected to the drive motor 14 to achieve smooth rotational movement inside the collection bucket 1.

[0055] The interior of the shell of the ball picking robot is further divided into a first storage area 10 and a second storage area 11. The second storage area 11 is arranged close to the transition portion 3, and is used to store the table tennis balls judged as bad balls by the ball picking robot; the first storage area 10 is arranged away from the transition portion 3, and is used to store the table tennis balls judged as good balls by the ball picking robot.

[0056] like Figure 3 and Figure 4As shown, two tracks 4 are arranged perpendicular to each other and intersecting at the ends inside the shell of the ball picking robot. One of the tracks 4 is arranged close to the transition part 3 and parallel to the transition part 3, and the track 4 is used to transport the table tennis balls dropped from the transition part 3 to the inside of the shell to the other track 4. The other track 4 is arranged perpendicular to the transition part 3 and passes through the second storage area 11 and the first storage area 10 in sequence. A camera 9 is arranged at the intersection of the two tracks 4, which is used to take the surface image of the table tennis balls and send it to the microcomputer 8, so as to judge the quality of the table tennis balls. A piston push rod 5 is arranged on the side of the track 4 arranged perpendicular to the transition part 3 away from the second storage area 11 in the width direction, and the piston push rod 5 is arranged directly opposite to the second storage area 11, and is used to push the bad balls into the second storage area 11. In order to ensure that the track 4 can transport the table tennis balls smoothly and prevent the table tennis balls from rolling around, a number of round holes can be opened on the track 4, and the diameter of the round holes is slightly smaller than the maximum outer diameter of the table tennis balls. A fan system 6 is provided on the same side of the piston push rod 5 as the first storage area 10, for blowing good balls into the first storage area 10. A fan system 6 is also provided at the bottom of the first storage area 10, for blowing good balls stored in the first storage area 10 into the base.

[0057] The ping-pong ball that falls on the crawler 4 arranged parallel to the transition part 3 is transported by the crawler 4 to the crawler 4 arranged perpendicular to the transition part 3, and then transported by the crawler 4 arranged perpendicular to the transition part 3 to the camera 9. The camera 9 takes an image of the ping-pong ball and sends it to the microcomputer 8. The image recognition system inside the microcomputer 8 distinguishes the quality of the ping-pong ball. If it is a bad ball, it sends a command to the piston push rod 5, and the piston push rod 5 pushes the bad ball into the second storage area 11; if it is a good ball, it sends a command to the fan system 6 to send the good ball to the first storage area 10. When the ball-picking robot collects the ping-pong balls and returns to the base, the fan system 6 at the bottom of the first storage area 10 starts to work, blowing the ping-pong ball upward into the pipe 18, and the ping-pong ball hits the baffle 17 and falls into the storage box 12.

[0058] A laser radar sensor 7 is also provided on the top of the shell of the ball picking robot and on the side connected to the transition part 3, which is used to send the collected data to the microcomputer 8. The microcomputer 8 processes the data sent by the laser radar sensor 7 through the internal path planning system. The laser radar sensor 7 constructs a detailed three-dimensional map of the surrounding environment by emitting a laser beam and accurately measuring the time of reflection. This process can not only identify the position of the table tennis ball, but also detect possible obstacles, such as the edge of the table, other balls or debris on the ground, to ensure the safety of the robot's actions. The microcomputer 8 receives the data from the laser radar sensor 7 and processes it at high speed. The path planning algorithm embedded in the microcomputer 8 will calculate the optimal ball picking path based on real-time environmental information. This path must not only ensure that all table tennis balls are picked up efficiently, but also take into account factors such as avoiding collisions and possible energy consumption. The microcomputer 8 includes the laser radar sensor 7 obstacle detection algorithm and the DynamicA* algorithm.

[0059] The present invention also designs a working method of the ball-picking split robot applicable to the present invention for intelligently identifying the quality of table tennis balls, comprising the following steps:

[0060] Step 1: The laser radar sensor 7 emits a laser beam, receives a reflected laser signal, and sends the laser signal to the microcomputer 8.

[0061] Step 2: The microcomputer 8 processes the laser signal to obtain the optimal path.

[0062] Step 21, the microcomputer 8 obtains the distance to each point in the environment according to the round-trip time of the laser beam and generates point cloud data.

[0063] Step 22: pre-process the point cloud data, perform point cloud registration, perform environmental modeling again, and construct a three-dimensional model of the environment surrounding the ball-picking robot.

[0064] The preprocessing of the collected point cloud data includes filtering and denoising to remove noise and abnormal points. The purpose of point cloud registration is to ensure that the point cloud data collected at different times can be accurately aligned.

[0065] Step 23: Detect and classify obstacles based on the three-dimensional model obtained in step 22 to distinguish between the ping-pong ball and the obstacles.

[0066] By analyzing the point cloud data, obstacles in the environment are identified, such as the edge of the table tennis table, other balls, or debris on the ground. The detected obstacles are classified to distinguish between the table tennis ball and other obstacles.

[0067] Step 24: Perform dynamic path planning according to the DynamicA* algorithm.

[0068] Step 241, initialize and define the starting point S, i.e. the initial position of the ball-picking robot, and the target point T, i.e. the position of the table tennis ball. Initialize the open list OpenList, add the starting point S to it, set g(S) = 0, f(S) = h(S), where h(S) is calculated according to the heuristic function, preferably using the Euclidean distance, the formula is as follows: Where (x s ,y s ) is the starting point coordinate, (x t ,y t ) is the target point coordinate. Initialize the closed list ClosedList, which is initially empty.

[0069] Step 242, calculate the evaluation function f(m) of each neighbor node m of node n. If node m is in ClosedList, skip the evaluation function calculation of this node, take the node m corresponding to the minimum value, put it into OpenList, put the node n originally in OpenList into ClosedList, and update the newly put node m to node n. Repeat this step until node n is the target point T. At this time, the path composed of all coordinates in ClosedList is the optimal path P. The calculation formula of the evaluation function f(m) is as follows:

[0070] f(m)=g(m)+h(m)

[0071] g(m)=g(n)+c*d

[0072] Where g(n) and g(m) are the actual costs from the starting point to nodes n and m respectively; h(m) is the heuristic estimated cost from node m to the target point; c is the cost coefficient, if the area where node m is located is an ordinary floor, the cost coefficient is c1, if the area where node m is located is cluttered, the cost coefficient is c2; d is the distance from node n to node m.

[0073] The node n has four neighbor nodes m, which are located above, below, left and right of the node n.

[0074] For each neighbor node m of node n: if m is in the ClosedList, skip it. h(m) is calculated based on the coordinates of the ball-picking robot and the table tennis ball, preferably using the Euclidean distance, as follows: ), where (x m ,y m ) are the coordinates of node m.

[0075] Step 243: If a new obstacle is detected at node k, update the cost of the affected node, take the node where the ball-picking robot is located as the initial node, repeat step 242, and obtain a new optimal path P2.

[0076] In the DynamicA* algorithm, the evaluation function f(n) of node n is defined as: f(n) = g(n) + h(n). g(n) represents the actual cost from the starting node to node n, such as the actual distance moved by the ball-picking robot. h(n) represents the heuristic estimated cost from node n to the target node, such as the Euclidean distance or Manhattan distance. For the ball-picking robot, in a dynamic environment, the calculation of g(n) needs to consider the movement cost of the ball-picking robot on different terrains, such as smooth floors, areas with debris, etc. Assume that the cost of the ball-picking robot moving a unit distance on an ordinary floor is c1, and the cost of moving a unit distance in an area with debris is c2. If the ball-picking robot passes through a series of nodes n1, n2,…, n from the starting point S. i Arrives at node n, and node n i The area to n is a debris area, the distance is d, then g(n) = g(n i )+c2d. The goal of the ball-picking robot is to reach the location of the ping-pong ball and pick it up. h(n) can be calculated based on the coordinates (x r ,y r ), the coordinates of the ball picking robot and the coordinates of the target point (x p ,y p ), i.e., the ping-pong ball coordinates. For example, using the Euclidean distance as a heuristic function When the environment changes, such as when new obstacles appear or the ping-pong ball is moved, the path needs to be recalculated. Suppose the planned path at time t1 is P1 = {n1, n2, ..., n m}, at time t2, a new obstacle is detected at node n k First, the g(n) and f(n) values ​​of the affected nodes need to be updated. The evaluation function of each node is recalculated with the position of the ball-picking robot as the initial value, and the cost of the new path segment is accumulated. The search is repeated to find the new path P2.

[0077] For a specific example, the ball-picking robot is located in a room with a normal floor area and some debris storage areas. The size of the room is 10m*10m. The initial position of the ball-picking robot is in one corner of the room (1,1). The ping-pong ball, that is, the target point, is in another corner of the room (9,9). In the room, there are some randomly placed debris. For example, the area from coordinates (3,3) to (4,4) is a debris storage area. Starting point S = (1,1), target point T = (9,9). Open list OpenList: add the starting point S to it. Calculate g(S) = 0, and calculate the heuristic function based on the Euclidean distance Then f(S)=h(S)≈11.31.

[0078] In each loop, search for the node n with the smallest f(n) value in OpenList. At the beginning, there is only the starting point S in OpenList, so n=S. Consider the neighbor node m of node n=S. The ball-picking robot can move in four directions, namely up, down, left, and right, and the moving distance in each direction is 1 unit. For the neighbor node m=(2,1), move one unit to the right, and the area where the node is located is an ordinary floor. Calculate the g value from the starting point to m through n: g(m)=g(n)+c1*d, where c1=1, d=1, so g(m)=0+1×1=1. Calculation The nodes m in the other three directions are calculated in the same way and compared. The node m corresponding to the minimum value is taken and put into the OpenList. The node n originally in the OpenList is put into the ClosedList, and the newly added node m is updated to the node n. Repeat this step until the node n is the target point T. At this time, the path composed of all the coordinates in the ClosedList is the optimal path P.

[0079] At time t2, a new obstacle is detected at node k = (5,5), and the previously planned path P passes through this node. Update the cost of the affected node, take the node where the ball-picking robot is located as the initial node, repeat the above steps, and get the new optimal path P2.

[0080] Step 3: The ball picking robot collects the table tennis balls according to the optimal path.

[0081] Step 4: When the collector 2 contacts the ping-pong ball, it starts to rotate and sends the ping-pong ball into the housing through the transition portion 3.

[0082] Step 5, the ping-pong ball that enters the shell falls on the track 4 arranged parallel to the transition portion 3, is transported by the track 4 to the track 4 arranged perpendicular to the transition portion 3, and then is transported to the camera 9 by the track 4 arranged perpendicular to the transition portion 3. The camera 9 takes an image of the ping-pong ball and sends it to the microcomputer 8.

[0083] Step 6, the microcomputer 8 distinguishes good balls from bad balls. If it is a bad ball, a command is sent to the piston push rod 5, and the piston push rod 5 pushes the bad ball into the second storage area 11; if it is a good ball, a command is sent to the fan system 6 to send the good ball into the first storage area 10.

[0084] Step 61, the microcomputer 8 extracts key features of the image, including color, texture, and shape, and preprocesses the image to obtain a preprocessed image;

[0085] Image preprocessing includes denoising, enhancement, color correction and image standardization. Denoising applies filters to reduce image noise, such as Gaussian filtering or median filtering. Enhancement adjusts the contrast and brightness of the image to highlight the details of the ping pong ball. Color correction ensures that the color of the ping pong ball is accurate in the image. Image standardization scales the image to a uniform size for subsequent processing. Image storage and management stores the preprocessed images in a database or file system of the microcomputer 8 for easy retrieval and management. Metadata is used to organize image data based on timestamps, camera parameters and image quality scores.

[0086] Step 62: Use the Canny edge detection algorithm to obtain the best edge detection result.

[0087] Step 621: For each small block of the preprocessed image, the local variance is calculated according to the three RGB channels respectively. The formula is as follows:

[0088]

[0089] Among them, r ij is the pixel value of the small block, i and j are integers greater than 0; is the average value; is the variance.

[0090] Step 622: According to the variance obtained in step 621, select the corresponding Gaussian kernel size.

[0091] For example, if the variance of the three RGB channels is greater than 100, a 5x5 Gaussian kernel is selected.

[0092] Step 623: Preliminarily detect the edge area in the image, adjust the pixels within 3 pixels from the edge to 0.5 times the original 3x3 Gaussian kernel weight, and obtain an image after adaptive Gaussian filtering.

[0093] Step 624: Use the Scharr operator to perform convolution operations on the image after the adaptive Gaussian filtering in the horizontal and vertical directions, respectively, to obtain the horizontal gradient component Gx and the vertical gradient component Gy.

[0094] Step 625: Select two scales. The small scale is the original image size, and the large scale is the original image downsampled to half its original size, that is, the width and height are both half of the original. Use the Scharr operator to calculate the gradient components Gx1 and Gy1 on the small scale image, and calculate Gx2 and Gy2 on the large scale image. Fuse the gradient information using a weighted average. The gradient amplitude after fusion is The formula for calculating the average gray value G and standard deviation σG of the entire table tennis image is as follows: Where gi is the grayscale value of the pixel in the image, N is the total number of pixels in the image,

[0095] Step 626: Set a high threshold Low Threshold Perform the first edge detection using a high threshold T h and low threshold T l , count the ratio of the number of detected edge pixels to the total number of pixels. If this ratio is less than 5%, it means that the edge is too sparse, then adjust the high threshold to T h =T h -5, the low threshold is adjusted to T l =T l -5, and then perform edge detection again; if this ratio is greater than 30%, it means that the edges are too dense, so adjust the high threshold to T h =T h +5, low threshold adjusted to T l =T l +5, perform edge detection again to get the best edge detection result.

[0096] Step 63, observe the edge contour obtained by the best edge detection result. If the edge contour is continuous and smooth, without obvious breakpoints or jagged shapes, and after fitting the edge contour, the deviation between the radius of the circle obtained by fitting and the radius of the ideal table tennis ball is less than 10%, then it is a good ball, and the microcomputer 8 sends a command to the fan system 6 to send the good ball to the first storage area 10; otherwise, it is a bad ball, and the microcomputer 8 sends a command to the piston push rod 5 to send the bad ball to the second storage area 11.

[0097] Observe the edge contour obtained by the best edge detection result. If the edge contour is continuous and smooth without obvious breakpoints or jaggedness, it means that the surface of the ping-pong ball is relatively intact without obvious damage or scratches. Fit the edge points obtained by edge detection and try to fit them into a circle. Calculate the deviation between the fitted circle and the ideal circle. Calculate the deviation between the radius of the fitted circle and the radius of the ideal ping-pong ball (the known standard radius is 20mm). If the deviation is less than 2mm (the radius deviation is less than 10%), it is considered that the shape of the ping-pong ball meets the requirements and is a ping-pong ball with a good shape; if the deviation is greater than 2mm, it may mean that the ping-pong ball has an irregular shape and poor quality.

[0098] The Canny edge detection algorithm includes noise suppression, calculation of gradient strength and direction, non-maximum suppression and dual threshold edge connection. Noise suppression uses a Gaussian filter to smooth the image to reduce the impact of noise. Calculation of gradient strength and direction applies operators to calculate the gradient strength and direction of each pixel in the image. Non-maximum suppression compares the pixel value with neighboring pixels along the gradient direction to ensure that only local maxima are marked as edges. Dual threshold edge connection sets two thresholds, high and low, the high threshold is used to determine strong edges, and the low threshold is used to connect edges. Any pixel with a gradient value above the high threshold is considered an edge, while pixels below the low threshold are suppressed. Pixels in between are considered edges if they are connected to a strong edge. Noise suppression can use Gaussian filtering to remove noise.

[0099] For example, using a 3x3 Gaussian kernel: the kernel matrix is ​​[[1 / 16,2 / 16,1 / 16],[2 / 16,4 / 16,2 / 16],[1 / 16,2 / 16,1 / 16]], which is the normalized result. The calculated gradient magnitude G and gradient direction θ, where G = sqrt(Gx^2+Gy^2), θ = arctan(Gy / Gx), Gx and Gy are the gradients in the x and y directions respectively.

[0100] The purpose of non-maximum suppression is to find the local maximum in the gradient amplitude image and set the gradient value corresponding to the non-maximum point to 0, thereby refining the edge. Along the gradient direction, compare the gradient amplitude of each pixel with its adjacent pixels. If the gradient amplitude of the pixel is not the local maximum, it is marked as a non-edge point. If the gradient amplitude of the pixel is greater than the high threshold, it is considered a strong edge point; if it is less than the low threshold, it is considered a non-edge point; if it is between the two, it is considered an edge point only when it is connected to a strong edge point.

[0101] For a specific example, take an image and get a color image in RGB format. For each small block, calculate the local variance using the three RGB channels. For the pixel value r in a small block of the red channel ij (i=1,…,8,j=1,…,8), the average value is The local variance is Similarly, calculate the local variance of the green and blue channels. Determine the Gaussian kernel size based on the calculated local variance. If the maximum local variance of a small block in the three channels is greater than 100, use 5x5 on the small block. Preliminary detection of edge areas in the image. For pixels near the edge, the range of less than or equal to 3 pixels from the edge pixel, adjust the weight during Gaussian filtering. Let the original 3x3 Gaussian kernel weight be w. For pixels near the edge, reduce the weight perpendicular to the edge direction to 0.5 times the original weight to better preserve edge information. For the image after adaptive Gaussian filtering, use the Scharr operator to perform convolution operations in the horizontal and vertical directions respectively to obtain the horizontal gradient component Gx and the vertical gradient component Gy. Select two scales, the small scale is the original image size, and the large scale is to downsample the original image to half of the original size, with the width and height both becoming half of the original. Use the Scharr operator to calculate the gradient components Gx1 and Gy1 on the small-scale image, and calculate Gx2 and Gy2 on the large-scale image. Fusion of gradient information, using weighted average. Gradient amplitude after fusion Calculate the average gray value of the entire ping pong ball image And standard deviation σG. For the fused gradient image, set it as a grayscale image, Where gi is the grayscale value of the pixel in the image, N is the total number of pixels in the image, Set a high threshold based on experience Low Threshold Perform the first edge detection using a high threshold T h and low threshold T l , count the ratio of the number of detected edge pixels to the total number of pixels. If this ratio is less than 5% (the edge is too sparse), adjust the high threshold to T h =T h -5, the low threshold is adjusted to T l =T l -5, and then perform edge detection again. If this ratio is greater than 30% (edges are too dense), adjust the high threshold to T h =T h +5, low threshold adjusted to T l =T l+5, perform edge detection again. Observe the edge contour obtained by the final edge detection result. If the edge contour is continuous and smooth, without obvious breakpoints or jaggedness, it means that the surface of the ping-pong ball is relatively intact, without obvious damage or scratches. Fit the edge points obtained by edge detection and try to fit it into a circle. Calculate the deviation between the fitted circle and the ideal circle. Calculate the deviation between the radius of the fitted circle and the radius of the ideal ping-pong ball. The standard radius is known to be 20mm. If the deviation is less than 2mm, that is, the radius deviation is less than 10%, it is considered that the shape of the ping-pong ball meets the requirements and is a well-shaped ping-pong ball; if the deviation is greater than 2mm, it may mean that the ping-pong ball has an irregular shape and poor quality.

[0102] Through Canny edge detection, we can detect irregular edges such as scratches and dents on the surface of a table tennis ball, thereby distinguishing the quality of the table tennis ball. Image analysis and understanding uses machine learning or deep learning models to analyze the extracted features to judge the quality of the table tennis ball. The deep learning architecture CNN is used in the training process, and the data set is divided into a training set, a validation set, and a test set. The training set is used to train the model, and the validation set is used to monitor the performance of the model to prevent overfitting. Hyperparameter adjustment, adjust hyperparameters such as learning rate, batch size, and number of network layers to obtain the best performance. Finally, the system outputs the quality assessment results of the table tennis ball, including information such as quality grade, defect type and location. The results are used to control the piston push rod 5 and the fan system 6 to classify good and bad balls.

[0103] Step 7, when the ping-pong balls in the ball picking robot are full or the power of the battery 15 is less than 10% of the total power, the ball picking robot returns to the base to charge; the fan system 6 at the bottom of the first storage area 10 works to blow the good balls in the first storage area 10 along the pipe 18 into the ball storage box 12 on the top of the base.

[0104] In addition to the above embodiments, the present invention may also have other implementation modes. Any technical solutions formed by equivalent replacement or equivalent transformation shall fall within the protection scope required by the present invention.

Claims

1. A split robot for picking up table tennis balls that can intelligently identify the quality of table tennis balls, characterized by: The invention comprises a base and a ball picking robot. The base is a box structure. The base is divided into three layers, which are a ball storage box (12), a storage box and a charging bin from top to bottom. The top of the ball storage box (12) is open and provided with a top cover. A pipe (18) is also provided inside the base and is vertically arranged and passes through the ball storage box (12), the storage box and the charging bin and is connected to the ball picking robot. The ball picking robot comprises a battery (15) arranged at the bottom, a universal wheel (13) arranged at the bottom of the battery (15), a shell arranged at the top of the battery (15), and a drive motor. (14) and a microcomputer (8); the ball picking robot also includes a collection bucket (1) arranged on one side of the battery (15), a collector (2) arranged inside the collection bucket (1) and rotatably connected to the collection bucket (1), a transition portion (3) arranged between the collection bucket (1) and the outer shell, two crawlers (4) arranged inside the outer shell and arranged perpendicularly to each other, a camera (9) arranged at the intersection of the two crawlers (4), and a laser radar sensor (7) arranged on the top of the outer shell and close to one side of the transition portion (3); the collection bucket (1) is away from the battery (15 ) is provided with an opening at one end; a first storage area (10) and a second storage area (11) are also provided inside the shell, and the second storage area (11) is provided near the transition part (3); one of the crawlers (4) is provided near the transition part (3) and parallel to the transition part (3), and the other crawler (4) is provided perpendicular to the transition part (3) and passes through the second storage area (11) and the first storage area (10) in sequence; a piston push rod (5) is provided on the side of the crawler (4) arranged perpendicular to the transition part (3) away from the second storage area (11) in the width direction, and the piston The push rod (5) is arranged opposite to the second storage area (11); a fan system (6) is arranged on the same side of the piston push rod (5) opposite to the first storage area (10), and the fan system (6) is also arranged at the bottom of the first storage area (10); the microcomputer (8) is electrically connected to the battery (15), and is connected to the drive motor (14), the fan system (6), the piston push rod (5), the camera (9) and the laser radar sensor (7) through signals; the drive motor (14) is connected to the universal wheel (13), the collector (2) and the crawler (4).

2. The ball-picking split robot for intelligently identifying the quality of table tennis balls according to claim 1, characterized in that: A baffle (17) is provided on the top of the pipeline (18).

3. The ball-picking split robot for intelligently identifying the quality of table tennis balls according to claim 1, characterized in that: The collector (2) is composed of a plurality of groups of blades arranged in an axial array along the collecting barrel (1), and each group of blades is composed of three blades arranged in a circumferential array.

4. The ball-picking split robot for intelligently identifying the quality of table tennis balls according to claim 3 is characterized by: The blade has a curvature that fits a table tennis ball.

5. The ball-picking split robot for intelligently identifying the quality of table tennis balls according to claim 1, characterized in that: The transition part (3) is wider at the portion connected to the collection bucket (1) and narrower at the portion connected to the outer shell, and the transition part (3) is tilted so that the portion connected to the collection bucket (1) is lower than the portion connected to the outer shell.

6. The ball-picking split robot for intelligently identifying the quality of table tennis balls according to claim 1, characterized in that: A plurality of circular holes are provided on the crawler belt (4).

7. The ball-picking split robot for intelligently identifying the quality of table tennis balls according to claim 6, characterized in that: The diameter of the circular hole formed on the crawler belt (4) is smaller than the maximum outer diameter of a table tennis ball.

8. The ball-picking split robot for intelligently identifying the quality of table tennis balls according to claim 1, characterized in that: A magnetic charger (16) is arranged at the charging compartment of the base, and the magnetic charger (16) cooperates with the battery (15).

9. A working method of a ball-picking split robot applicable to any one of claims 1 to 8 for intelligently identifying the quality of table tennis balls, characterized in that: The following steps are included: S1, the laser radar sensor (7) emits a laser beam, receives a reflected laser signal, and sends the laser signal to a microcomputer (8); S2, a microcomputer (8) processes the laser signal to obtain an optimal path; S21, the microcomputer (8) obtains the distance to each point in the environment according to the round-trip time of the laser beam and generates point cloud data; S22, preprocessing the point cloud data, performing point cloud registration, and performing environmental modeling again to construct a three-dimensional model of the environment surrounding the ball picking robot; S23, detecting and classifying obstacles according to the three-dimensional model obtained in step S22, and distinguishing between table tennis balls and obstacles; S24, performing dynamic path planning according to the DynamicA* algorithm; S241, initialize and define the starting point S and the target point T, initialize the open list OpenList, add the starting point S to it, set g(S) = 0, f(S) = h(S), where h(S) is calculated according to the heuristic function, initialize the closed list ClosedList, which is initially empty; S242, calculate the evaluation function f(m) of each neighbor node m of node n. If node m is in ClosedList, skip the evaluation function calculation of this node, take the node m corresponding to the minimum value, put it into OpenList, put the node n originally in OpenList into ClosedList, and update the newly put node m to node n. Repeat this step until node n is the target point T. At this time, the path composed of all coordinates in ClosedList is the optimal path P. The calculation formula of the evaluation function f(m) is as follows: f(m)=g(m)+h(m) g(m)=g(n)+c*d Where g(n) and g(m) are the actual costs from the starting point to nodes n and m respectively; h(m) is the heuristic estimated cost from node m to the target point; c is the cost coefficient, if the area where node m is located is an ordinary floor, the cost coefficient is c1, if the area where node m is located is cluttered, the cost coefficient is c2; d is the distance from node n to node m; S243, if a new obstacle is detected at node k, update the cost of the affected node, take the node where the ball-picking robot is located as the initial node, repeat step S242, and obtain a new optimal path P2; S3, the ball picking robot performs table tennis ball collection according to the optimal path; S4, when the collector (2) contacts the ping-pong ball, it starts to rotate, and sends the ping-pong ball into the housing through the transition portion (3); S5, the table tennis ball that has entered the housing falls on a crawler (4) arranged parallel to the transition portion (3), is transported by the crawler (4) to a crawler (4) arranged perpendicular to the transition portion (3), and is then transported by the crawler (4) arranged perpendicular to the transition portion (3) to a camera (9), where the camera (9) captures an image of the table tennis ball and sends the image to a microcomputer (8); S6, the microcomputer (8) distinguishes good balls from bad balls. If the ball is a bad ball, a command is sent to the piston push rod (5), and the piston push rod (5) pushes the bad ball into the second storage area (11); if the ball is a good ball, a command is sent to the fan system (6) to send the good ball into the first storage area (10); S61, the microcomputer (8) extracts key features from the image and preprocesses the image to obtain a preprocessed image; S62, using the Canny edge detection algorithm to obtain the best edge detection result; S621, for each small block of the preprocessed image, calculate the local variance according to the three RGB channels respectively, the formula is as follows: Among them, r ij is the pixel value of the small block, i and j are integers greater than 0; is the average value; is the variance; S622, selecting a corresponding Gaussian kernel size according to the variance obtained in step S621; S623, preliminarily detecting the edge area in the image, adjusting the pixels within 3 pixels from the edge to 0.5 times the weight of the original 3x3 Gaussian kernel, and obtaining an image after adaptive Gaussian filtering; S624, performing convolution operations on the image after adaptive Gaussian filtering in the horizontal and vertical directions using the Scharr operator to obtain a horizontal gradient component Gx and a vertical gradient component Gy; S625, select two scales, the small scale is the original image size, the large scale is the original image downsampled to half of the original size, use the Scharr operator on the small scale image to calculate the gradient components Gx1, Gy1, and calculate Gx2, Gy2 on the large scale image; fuse the gradient information, using the weighted average method; the gradient amplitude after fusion Calculate the average gray value of the entire ping pong ball image And the standard deviation σG formula is as follows: where g i is the grayscale value of the pixel in the image, N is the total number of pixels in the image, S626, set high threshold Low Threshold Perform the first edge detection and count the ratio of the number of edge pixels detected to the total number of pixels; if the ratio is less than 5%, adjust the high threshold to T h =T h -5, the low threshold is adjusted to T l =T l -5; if the ratio is greater than 30%, the high threshold is adjusted to T h =T h +5, low threshold adjusted to T l =T l +5, perform edge detection again to obtain the best edge detection result; S63, observing the edge contour obtained by the best edge detection result, if the edge contour is continuous and smooth without obvious breakpoints or jagged shapes, and after fitting the edge contour, the deviation between the radius of the circle obtained by fitting and the radius of the ideal table tennis ball is less than 10%, then it is a good ball, and the microcomputer (8) sends a command to the fan system (6) to send the good ball to the first storage area (10); otherwise, it is a bad ball, and the microcomputer (8) sends a command to the piston push rod (5) to send the bad ball to the second storage area (11); S7. When the ball-picking robot is fully loaded with table tennis balls or the power level of the battery (15) is less than 10% of the total power level, the ball-picking robot returns to the base to charge; the fan system (6) at the bottom of the first storage area (10) starts to work, blowing the good balls in the first storage area (10) into the ball storage box (12) along the pipeline (18).

10. The working method of the ball-picking split robot for intelligently identifying the quality of table tennis balls according to claim 9, characterized in that: The heuristic estimation cost function h(*) from the calculation node to the target point in steps S241 and S242 is a Euclidean distance function.

Images