Abstract: Based on the behavior of an improved visual recognition algorithm,and applied to the bus staff dangerous motion detection,based on visual identification results of target detection using dynamic shape of the 3ddata describing the filtered structural elements,build the bus staff dangerous action area,the outline of base according to their shape characteristics compared 3dbehavior characteristic database related personnel,personnel risk action implementation bus intelligent detection.The simulation results show that the proposed method performs dangerous behavior of the personnel on the car for testing,the result has higher accuracy and effectiveness.