Application of improved target detection algorithm in sports robot motion behavior

Abstract

Target detection in classroom education scene often brings some difficulties to target detection based on YOLO due to the large detection range and small detection target in classroom. In this study, target detection methods DPM and R-FCN were integrated into YOLO and an improved neural network structure was designed. The feature extraction mode included a fully connected layer and pooling and then convolution to reduce the loss of feature information. Then, a sliding window merging algorithm based on RPN was designed to form a feature extraction algorithm based on improved YOLO. In this study, a context detection platform for educational robot was built to clarify the overall workflow of context detection for educational robot. the comparison with the YOLO algorithm shows that the proposed algorithm is superior to the YOLO algorithm in recognition accuracy.