Advances in Computer Science Research, volume 89
2019 International Conference on Wireless Communication, Network and Multimedia Engineering (WCNME 2019)
Intelligent Tennis Picking Robot Based on STM32 and
Raspberry Pi
Shiqian Zhang1, Yizhun Peng2,*, Zhenjiang Chen, Di Li and Zhou Zhang
1
College of Electronic Information and Automation, Tianjin University of Science and Technology, China
2
Advanced Structural Integrity International Joint Research Centre, Tianjin University of Science and Technology, Tianjin, 300222,
China
*
Corresponding author
Abstract—The robot involves a smart tennis picking robot personnel is poor. Most of them are temporary employment of
based on STM32 and Raspberry Pi. The mobile robot body is tennis courts[2]. Therefore, designing such an intelligent tennis
formed by sheet metal forming. The driving method is three-wheel picking robot based on omnidirectional mobile platform is of
independent driving, and three hollow cup DC servo motors are great significance for tennis training. The introduction of this
respectively controlled by three. An omnidirectional wheel. The intelligent robotic product on the tennis court not only saves
robot is divided into a chassis motion control portion, a visual money, but also satisfies the pursuit of fashion for tennis lovers.
processing portion, and a tennis collection portion. The chassis
control section is located at the bottom of the robot, including the
servo drive, battery, main control board, HC05 Bluetooth module, II. MECHANICAL STRUCTURE AND CONTROL HARDWARE
two ultrasonic obstacle avoidance sensors and three hollow cup The smart tennis picking robot chassis is formed by sheet
DC servo motors. The visual processing part consists of a metal forming, and the outer dimensions are 365.5×400.2×109.
raspberry pie and a camera. The tennis collection section includes The wheels used are omnidirectional wheels. And the overall
a tennis collection structure and a storage box. The STM32 chassis structure is shown in Figure 1.
controller interacts with the servo drive via the CAN bus, enabling
the robot to move omnidirectionally with zero radius of rotation.
At the same time, the Raspberry-based Haar-based image
recognition program and camera can accurately identify the
tennis target position. Ingenious science solves the problem of time
and labor for manual tennis picking.
This paper introduces the overall mechanical structure of the
robot, describes the mechanical structure of the chassis and tennis
collection part and the advantages and reasons for selecting the
robot control hardware. Starting from the image recognition part,
the tennis recognition processing program combining Haar
classifier and CamShift algorithm is described in detail by
modularization. Finally, the advantages of the design product and
the future development of other functions of the robot are
summarized.
Keywords—tennis picking; raspberry Pi; STM32; haar; FIGURE I. CHASSIS STRUCTURE
CamShift
The use of such a chassis structure makes the robot more
flexible and can achieve three degrees of freedom (horizontal,
I. INTRODUCTION
vertical and rotational) motion in either plane[3]. The maximum
Nowadays, more and more people are engaged in the army moving speed of the robot is 0.65m/s, the maximum rotation
of sports. Tennis is a very popular competitive sport and attracts speed is 200°/s, and the minimum turning radius is 0m. Due to
a large number of fans. Because the tennis court has a large area its low ground, it is suitable for indoor and outdoor tennis courts.
and a fast ball speed, people also consume a lot of physical The battery life is about 10h and the standby time is more than
strength because of the running. Therefore, it is a very painful 30h.
thing for tennis players to smash tennis. At present, there is no
service or product specializing in intelligent tennis picking At the front of the robot is a tennis collection structure
robots in China, which belongs to the market to be developed. consisting of a collection of forearms, tracks and brushes. The
At the same time, the Chinese market has great potential. There forearm is collected at the foremost end of the collecting
are about 85,475 tennis courts and 8.64 million tennis players, structure, followed by the ascending track, and a brush and a
and the number of sports people is growing at an annual rate of motor are fixedly mounted on the upper side of the ascending
10% to 12%[1]. At present, there are a small number of caddies track for moving the tennis ball from the track to the tennis
specializing in tennis croquet on the market, and the quality of receiving case on the upper side of the robot.
Copyright © 2019, the Authors. Published by Atlantis Press.
This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/). 195
Advances in Computer Science Research, volume 89
When collecting tennis balls, first collect the forearms to advances to the tennis ball, and the tennis picking mechanism
increase the collection range and improve the collection picks up the tennis ball into the collecting box. The program
efficiency. At the same time, the collected tennis balls are fed flow is shown in Figure 4.
onto the track by the rotating brush driven by the motor, thereby
entering the tennis receiving box. The receiving box and the
chassis control layer are separated by a laterally mounted
partition and fixed to the tennis collecting structure by a groove
for easy disassembly. The tennis collection structure is shown in
Figure 2.
FIGURE II. TENNIS COLLECTION STRUCTURE
The STM32f407 and an ARM-based servo drive are used to
control the chassis motion in terms of control hardware. The
STM32f407 is a high-performance 32-bit processor with a
168MHz clock speed and 210DMIPS processing power[4]. At
the same time, with the servo driver used as the control core of
ARM, complex control algorithms can be realized, which
reduces the volume of the system and increases the reliability of
the system. The servo driver workflow is shown in Figure3.
The control core used in image processing is the Raspberry FIGURE III. SERVO DRIVE WORKFLOW
Pi. Run the opencv software on the Linux system it is running on.
Low power consumption, small size, low cost and high
computing power make it an important reason for this robot
image recognition processing center[5,6].
III. ALGORITHM FLOW
The robot uses the Haar Cascade Classifier combined with
the CamShift algorithm to realize the recognition and tracking
function of the tennis ball. The CamShift algorithm can quickly
and efficiently track objects with fast target changes in the video,
with accurate positioning, small algorithm complexity and good
robustness[7]. After analyzing and comparing with several other
algorithms, it is determined that the CamShift algorithm is used
to realize the detection and recognition of tennis balls based on
the dynamic target color histogram[8].
The Haar Cascade Classifier is a supervised learning
classifier. To detect an object, first perform histogram
equalization on the image and normalize it to the same size[9],
FIGURE IV. PROCEDURE FLOW CHART
and then mark whether it contains the object to be detected.
When the image captured by the camera contains tennis, the
robot recognizes the tennis ball and executes CamShift. The IV. SUMMARY
algorithm calculates the deviation direction of the tennis ball
from the video center, thereby controlling the steering of the In view of the continuous popularization of tennis and the
robot. When reaching the center line of the screen[10], the robot main problems in daily tennis, this paper mainly designs a smart
196
Advances in Computer Science Research, volume 89
tennis picking robot, which is designed from the mechanical
structure and hardware selection to the system program. Finally,
the design of the whole robot is completed. The main function of
the structure is to pick up and collect the tennis balls scattered in
the tennis court by the robot, and solve the problem of time and
labor for manual tennis picking.
The robot has the following features: First, the chassis
motion control with high flexibility and high accuracy, the
omnidirectional movement and the turning radius are zero
through the omnidirectional wheel and the mechanical structure.
The second is to separate the chassis motion control from the
image recognition processing, and use STM32 and Raspberry Pi
as the control centers of these two parts respectively to realize
the separation of decision and control, make the system more
stable and reliable, and develop more convenient. Third, the
tennis pick-up mechanism is exquisitely designed and efficient,
and can meet the normal needs of users.
REFERENCES
[1] Chuan Xiao. Exploring the Reasons for the Rapid Development of Tennis
in China[J]. Thinking Times, 2019(05):243-244.
[2] Li-shan Zhang, Yu-feng Yuan. Research on the Formation and
Development of Chinese Tennis Culture[J]. Sports Excellence, 2018,
37(12): 46-47.
[3] Xiang-nan Fu. Research on dynamic modeling and control of
omnidirectional mobile robot [D]. Shenyang University of Technology,
2018.
[4] Yuan-zhen Su, Lei Ji, Jia-xin Nie, Guo-jiang Wei, Yu-qin Guan. Design
of Path Recognition Intelligent Car Based on STM32F407[J]. Information
& Communications, 2017(08):31-33.
[5] Zhen-jun Cao, Jun-feng Jing, Ze-bin Su, Slow Zhang. Research on the
depth detection of color fabric based on raspberry pie[J]. Cotton Textile
Technology, 2019, 47(01): 11-15.
[6] Hao You, Xin-xin Zhang, Zhao-hui Zhang. Motion detection based on
OpenCV and Raspberry Pi[J]. Electronic Technology, 2017, 46(02):
59-61.
[7] Ya-wen Li, Jie Zhao, Xin Zhang. Study on Moving Target Tracking
Algorithm Based on CamShift in Surveillance Video[J]. Ship Electronic
Engineering,2018,38(09):27-30.
[8] Shu-jing Zhang. Research on target tracking method of mobile robot
based on Camshift algorithm [D]. Beijing University of Civil Engineering
and Architecture, 2018.
[9] Si-yu Lin. Research on Fatigue Driving System Based on OpenCV[A].
Advanced Science and Industry Research Center.Proceedings of 2017 2nd
International Conference on Computer, Mechatronics and Electronic
Engineering(CMEE 2017)[C].Advanced Science and Industry Research
Center:Science and Engineering Research Center,2017:5.
[10] Ru Li, Jing-wen Hu. CamShift algorithm and embedded target tracking
system[J].Wireless Interconnect Technology,2018,15(18):76-77.
197
