Tennis Court Mapping and Trajectory Optimization for Robotic Object Retrieval

This paper presents a real-time tennis ball retrieval system, using computer vision and a blind LEGO MINDSTORMS EV3 robot. Our system tackles the tennis ball retrieval issue of lacking ball boys, during professional or amateur matches. This system is able to identify the location of a tennis ball in the whole tennis court, to drive a robot along a fast trajectory while avoiding collisions with players. Semantic information on the court is captured using a webcam, which is located randomly near the external border of the tennis court with a flexible height. The image is then processed using a mapping algorithm to correlate pixels in the image and points on the actual tennis court. The camera location is evaluated by comparing the lines of the tennis court in reality and in the image. Feature extraction and computer vision techniques are used to detect the tennis ball and the player, which are found with margins of 4.5 cm and 45 cm respectively. To optimize the ball retrieval process, a genetic algorithm is applied to define the optimal trajectory with fulfilling multiple objectives for the robot to travel, arrive and retrieve a tennis ball. MATLAB is used to process image extraction and processing, to evaluate the mapping algorithm, and to define the trajectory optimization on the image of the tennis court.