Study on the Segmentation Method of the Improved DeepLabv3+ Algorithm in the Basketball Scene

Abstract

Aiming at the weak robustness and low segmentation accuracy of traditional basketball scene segmentation methods, this study proposed an improved basketball scene semantic segmentation model based on DeepLabv3+ for the purpose of basketball scene segmentation and accurate positioning of players. In this model, a relatively complex decoder is designed based on the DeepLabv3+ network. Multiple feature fusion is used to restore the semantic information of the image, and the convolutional block attention mechanism is introduced to optimize the channel weight and position information, reduce the computational complexity of the model, and improve the edge sensitivity. Experimental results show that the proposed model is 21.8% better than FCN’s full convolution model and 1.9% better than DeepLabv3+. At the speed of segmentation, it can process 6 pictures per second, greatly improving the accuracy of semantic segmentation for basketball scenes. In the future, real-time detection of sports such as basketball using computer vision methods will become more and more important.