An Improved NanoDet-Based Multi-Human Detection Algorithm for Complex Motion Scenes
In sports event scenarios,athlete behaviour recognition,passing and shooting action count and AI commentary are inseparable from human body detection of athletes,which makes high requirements on the speed and accuracy of human body detection of athletes in complex scenarios.Therefore,this paper proposes a NanoDet-based multi-body detection algorithm for complex sports scenes.First,the algorithm uses a smoother Mish function as the activation function of the backbone network,improves the ShuffleNetV2 network,builds the backbone network,and introduces the CBAM attention module,and uses a lightweight path pooling network for feature fusion to improve detection accuracy;next,it uses the anchorless lightweight detection head GFLV2 for regression and classification to achieve multi-body target detection in complex motion scenes.Finally,in order to further verify the performance of the proposed algorithm,the research algorithm is experimentally compared with the current mainstream detection algorithms,the experimental results show that the algorithm in this paper has higher detection accuracy compared to other algorithms such as Yolov3-tiny and Yolov4-tiny,which is 14.87%higher than the same type of lightweight detection model Yolov4-tiny.In addition,the single-frame detection time is reduced by 31.68%compared to the 10.67ms of Yolov4-tiny,which shows that the investigated method substantially improves the detection accuracy while maintaining the detection speed improvement.
deep learninghuman detectionlightweight modelattention mechanism
万方数据
维普
