The Gesture Recognition Improvement of Mediapipe Model Based on Historical Trajectory Assist Tracking, Kalman Filtering and Smooth Filtering

With the rapid development of artificial intelligence and computer technology, how can computers be made simpler to operate, easier to learn, more adaptable to the natural state of activity of the operators, and accurately identify and understand them is an important and long-term research direction in the field of human-computer interaction. Gesture command recognition, as a popular direction in the field of human-computer interaction, has many solutions, among which the most famous is Mediapipe. However, in the current recognition and detection technology, there are still some unsatisfactory aspects of gesture recognition based on Mediapipe. This article focuses on the gesture recognition model of Mediapipe and improves and studies the layer crossing caused by occlusions of the same class as the identified target. The model is processed using historical trajectory assisted tracking and smooth filtering, and ablation experiments are conducted to confirm the effectiveness of the proposed approach. This article mainly uses Python 3.9 and PyCharm Professional Edition as the main tools for programming and experimentation.