Motion recognition method based on multi-model sequence fusion

The invention discloses a motion recognition method based on multi-model sequence fusion. The motion recognition method comprises the following steps: acquiring information of initial videos in various modes, pre-processing the acquired information to acquire an initial RGB image sequence and an initial depth image sequence, and also acquiring a framework feature sequence; carrying out background modeling on the initial RGB image sequence, extracting a first human body zone external rectangle according to the acquired background modeling result, and extracting a second human body zone external rectangle at the corresponding position of the initial depth image sequence; respectively extracting human body behavior features on the first and second human body zone external rectangles to obtain feature vectors; realizing behavior recognition through a multi-view distinguishing model according to the obtained RGB-LBP and D-LBP feature vector sequence and the framework feature sequence. According to the invention, sequence information of various models is complementarily fused, so that the behavior recognition accuracy can be effectively improved.