Unsafe Operation Recognition Based On The Depth Information Of The Driver

In recent years, the safe situation of traffics is very severe in our country. Traffic accidents and casualties caused buses are very serious in densely-populated cities every year. Nowadays, there are calling, hands off the steering wheel and fatigue driving behaviors of the bus driver happen occasionally. The existing major recognition method is that the inspector view supervision video system installed in the bus to check the driver’s unsafe operation behaviors or potential safety hazards after the bus stops. It has a lot of work, a high cost and a low real-time performance. We depend on image’s color and depth data obtained by Kinect designing a method to recognize calling, hands off the steering wheel, turning around and looking up and down unsafe operation behaviors during the driving process.The main work is as follows:Recognition algorithm about the driver’s operation behaviors with hands off steering wheel and calling based on depth information was proposed. Firstly, the algorithm adopted the background difference method to detect the driver’s area. Then it used bone locating method based on the depth information to find the driver’s hands and head region. And the skeleton movement information was analyzed to locate the driver’s hands movement. The algorithm achieved driver precise positioning depending on skin color model at the same time. Finally it identified unsafe operation behaviors of the driver’s about calling or hands off steering wheel. Experimental results certified that the method of bone and skin color model can avoid shakes for overlapping bone joint nodes when the bone positioned, and weakened the false detection effects caused external conditions such as light intensity when skin color detected.The method of bone and skin color model improves the recognition accuracy.Recognition algorithm about the driver’s unsafe behaviors with turning around and looking up and down based on depth information during the driving process was designed. Firstly, the algorithm used the skeleton and skin color method to detect the head area of the driver. Then it adopted combined depth image, color image with AAM algorithm to obtain facial feature points from 3D face information. And then the movement direction of the nose and eyes was analyzed to judge the head movements. Finally, it identified the driver’s unsafe behavior of turning around and looking up and down. Experimental results certified that this method has good robustness and real-time performance, and can achieve the desired effects.