满足RoboCup规范小型类人足球机器人设计与目标识别

Design of kid size humanoid soccer robot meeting the RoboCup specification and its object detection

设计了符合RoboCup技术规范的小型类足球机器人,并通过深度学习算法YOLO对比赛场地的目标进行训练和识别。分析了RoboCup技术规范对小型类人机器人设计的要求,设计并加工装配了满足要求的机器人。应用3000幅现场图片对深度学习神经网络进行训练,并对300幅不同于训练样本的图片进行识别,识别准确率为93. 8%,召回率为89. 7%,检测速度可达40 f/s。实验结果表明深度学习算法YOLO相比于其它算法应用于RoboCup足球机器人具有较好的鲁棒性和实时性。

A kid size soccer humanoid robot which meets RoboCup technical specification is designed, and the deep learning algorithm YOLO is trained to recognize the targets of the match field. The requirements of RoboCup technical specification for the design of small humanoid robot are analyzed, and the robot that meets the requirements is designed, processed and assembled. YOLO is trained with 3000 pictures of field objects and other 300 different pictures from training ones are identified. The experimental results show that the accuracy of the algorithm is 93.8%;the recall rate is 89.7%, and the detection speed can reach 40 frames per second. It shows that the deep learning algorithm YOLO has better robustness and real-time performance than other algorithms applied to RoboCup soccer robot in object detection.