A Novel Gesture Recognition Model Under Sports Scenarios Based on Kalman Filtering and YOLOv5 Algorithm
Tingting Wu, Xingfeng Fan · IEEE Access · 2024
With the development of computer vision, automatic gesture recognition towards sports scenarios has been more significant in recent years. However, due to the complexity of motion scenes and the diversity of target objects, existing gesture recognition methods still face certain challenges in terms of fine-grained feature perception. To deal with this issue, this paper proposes a novel gesture recognition model under sports scenarios using Kalman filtering theory and YOLOv5 Algorithm. Firstly, the Kalman filtering algorithm is used to preprocess attitude data of targets. In particular, it can optimize the attitude data in time series by combining sensor measurements and system models. Thus, motion state decoding is completed for timed updates of trajectories. Then, the object detection algorithm YOLOv5 is introduced to detect gestures of humans. In this part, the initial YOLOv5 algorithm is lightly improved by introducing lightweight backbone structure, in order to improve both detection efficiency and running efficiency. Finally, the Kalman filtering part is combined with YOLOv5 algorithm part to construct a comprehensive gesture recognition model under sports scenarios. After that, some real-world images of sports scenarios are utilized as the experimental scene to testify performance of the proposed method. The results show that it has advantage in accuracy, stability, and real-time performance by comparing with typical models.