Human Action Recognition Algorithm Based on DBPSO-SVM Classifier
Yunkun Ning, Sheng Zhang, Weimin Xiong, Guanglin Li, Guoru Zhao · 2019
In the context of population aging, the study of behavior recognition algorithms for the elderly has important social significance. This algorithm is designed to distinguish between Fall Action and Activity in Daily Life (ADL), which is important for the fall protection of older people. This paper makes full use of the improved Discrete Binary Particle Swarm Optimization (DBPSO) to select the optimal feature subset and parameters to train SVM (Support Vector Machine) algorithm. Experimental results show that this method can achieve higher precision than non-optimized SVM, the improved DBPSO-SVM algorithm can achieve more than 95% specificity and sensitivity for each action, especially when it comes to falls, the sensitivity is 98.44%, the specificity is 96.25%, and compared with the accuracy of 94.17% of the non-optimized SVM algorithm, there are good improvements.