Motion Posture Recognition Method Based on RBF Neural Network Optimization
Lezhong Sun · 2022
In order to improve the recognition rate of moving targets by radial basis function (RBF) neural network, a posture recognition algorithm based on optimized RBF is proposed. Firstly, the basic technology of human behavior recognition acquires in-depth analysis. Then, a dynamic adjustment of the important parameters of Levenberg-Marquardt with the number of iterative steps is put forward, to construct an adaptive RBF structure. Combined with the excellent global search ability of GA, it is also improved accordingly. Finally, the normalized motion history image (MHI) is used for image sequence representation for data processing. The simulation results show that our method reduces the over fitting degree of RBF, improves the generalization ability and calibration accuracy of LM-BP network, and can effectively recognize human posture categories.