Multi-person gesture recognition for complex environmental based on improved Yolov8 algorithm

Multi person gesture recognition in complex environmental is one of the branches of gesture recognition development, facing the problems of incomplete recognition and low recognition rates, due to high scene complexity, multiple recognition targets, and varying target sizes. In this paper, we based on the yolov8 algorithm, adding the GAM (Global Attention Mechanism), and integrating the GSConv (Group Shuffle Convolution) and Slim Neck to improve the yolov8 algorithm, combining the image segmentation technology with the SR algorithm Real-ESRGAN to form the YOSRLO model, which is aimed at providing a solution for the multi-person hand gesture recognition. We compare the experimental results of YOSRLO with yolov8, demonstrating that the YOSRLO model is highly accuracy, performs well, and can recognize small target gestures on Kinect and other camera devices.