Research on Multi-Sensor Combination Optimization for Skiing Behavior Recognition

This paper presents a multi-sensor combination optimization algorithm based on the discrete particle swarm algorithm for addressing the multi-sensor redundancy problem in human skiing behavior recognition. A combination of sensor locations with better recognition and fewer sensor locations is selected from multiple sensor locations. The fitness function design is rooted in the concept of minimizing redundancy while maximizing correlation, considering both sensor correlations and redundancy, distinguishing it from previous studies. To evaluate the effectiveness of the chosen combination of sensor locations, acceleration data from 12 body part sensors during four basic snowboarding maneuvers were collected and classified on four classifiers: KNN, SVM, CNN and AlexNet. The experimental results indicate that the location combination of six sensors located at the left calf, right thigh, right calf, left thigh, hip and back achieves the highest effectiveness, with a recognition accuracy of 89.24%. Furthermore, a comparison with the RF method, which only considers correlation, demonstrates the effectiveness of the multi-sensor combination optimization algorithm designed in this study.