A Gesture Recognition Method Based on Improved YOLOv8 Detection
Haiming Li, Yanfei Chen, Lei Yang, Jiaoxing Shi, Tao Liu, Yihui Zhou · 2024
Aiming at the existing hand gesture estimation algorithms with large computation and poor real-time performance, this paper proposes an improved YOLOv8 algorithm for hand recognition, which combines the improved YOLOv8 detection algorithm for the detection of the hand and combines it with gesture analysis to realize gesture recognition. The improved algorithm not only reduces the amount of floating-point arithmetic, but also increases the accuracy and robustness of detection. In the gesture detection part, the traditional YOLOv8 algorithm is improved by firstly adding the p6 detection head, secondly replacing the CBS module with RFSE module, and thirdly replacing the C2f module with CFE module. The gesture analysis part is performed by training the yolov8 classification model for gesture recognition. The test results of the improved model in TV-Hand and COCO-Hand datasets confirm the effectiveness of the improved YOLOv8 algorithm in this paper, which improves the accuracy by 2.3% and reduces the computational volume by 13.5% compared to the original YOLOv8n, and the correct rates of gesture recognition are all over 90%, and the improved algorithm, which ensures the lightweight and at the same time realizes the further enhancement of accuracy The improved algorithm can realize further improvement of accuracy while ensuring lightweight, and the application prospect is broader.