Study of Improved Yolov5 Algorithms for Gesture Recognition

As a new way of human-computer interaction, gesture recognition is an important means of communication for special people. It is widely used in virtual reality system. Because yolov5 network has the advantages of high detection accuracy and high speed, it has been widely used in the field of target detection. Now it is improved based on yolov5 network, which improves the learning and presentation ability of the network by modifying the convolution layer, integrating the attention mechanism of Se channel, image enhancement, data set processing and other methods, and adds Media Pipe key point detection technology at the input to improve the generalization ability of the model. The experimental results show that compared with the improved yolov5 model, the improved yolov5 model improves the harmonic average F1 score of accuracy and recall by about 20%, recall by about 7.2%, precision by about 6.8% and average recognition accuracy MAP by about 10
^%
on the premise that the detection speed remains basically unchanged. The improved yolov5 model can better meet the requirements of real-time and accuracy in the field of gesture recognition, and can provide important theoretical basis and reference value for transplanting to portable devices and the development of human-computer interaction technology in the future.