Research of Gesture Recognition Algorithm Based on Acceleration Trajectory Image

According to the characteristics of neural network computing, this paper designs a neural network based on acceleration for gesture detection. First, the acceleration information is collected by using the acceleration sensor to extract the key point information, and convert the effective data into the acceleration track image data. Two neural networks with depth distribution of 50 and 101 are built and trained by ResNet algorithm. Matching image data features to obtain models with accuracy rates of 85% and 91% respectively, so as to achieve higher gesture recognition accuracy with less computational time and storage space complexity.