Multi-person location and tracking method based on BP neural network
Pan Wei, Liu Zhizhan, Yi Zou · 2008
This paper focuses on the study of multi-person location and tracking in a complex scene created by 3ds max. To establish the complicated relationship between the 2D-image information that is obtained through the three-camera system and the 3D information of the target, an artificial neural network is proposed. In order to overcome the shortcomings of traditional BP algorithm as being slow to converge and easy to reach extreme minimum value, the model adopts LM (Levenberg-Marquardt) algorithm to achieve a higher speed and a lower error rate. Experiment results verify that the BP neural network improves the efficiency, accuracy and robustness of the method comparing with traditional binocular location (TBL) methods.