Detecting Driver Use Of Mobile Phone Based On Computer Vision

The driver use of mobile phone while driving can seriously interfere with the driver’s attention and the risk of occurring accident is four times higher than normal driving. Now, the study in using a mobile phone while driving is still limited, and most of them focus on the detection method based on the signal. However, the approach based on signal will generate a lot of false positives, because it is difficult to distinguish the driver or the passenger in a car who is using the phone. Therefore, our paper studies the driver use of mobile phone while driving in computer vision aspect. We proposed two different scenarios applied to the driver use of mobile phone, one is detecting driver use of mobile based on in-car camera, the other one is based on out-of-car camera.For detecting driver use of mobile phone based on in-car camera, we use the camera that is mounted on the windshield to capture driver’s video data and detect the phoning action by them. First, we decompose the phoning action into a series of atomic actions that conform to certain temporal relationships. Then, we use the And-or graph to model the hierarchy construction and relationships within atomics of phoning action. Finally, we detect the phoning action in our video data through statistical analysis methods. This method could be used as a supplement of the safety device inside the public vehicles, such as taxi, bus, truck, train and so on. If the driver is phoning while driving, the system will alert the driver, this can improve the traffic safety. For detecting driver use of mobile phone based on out-of-car camera, we use the camera that is mounted in the crossing or the overbridge to capture the driver’s image data. Detection under this scenario is a picture-based method. We judge the phoning action by detecting driver’s phoning action and hand’s state(hand located in the lower part of the head and hand is in the grabbing state). This method can be used for the transport sector to collect evidences for the punishment of driver’s phoning action.We bulid the video and image dataset of driver’s phoning action. Experimental results in our dataset show that the proposed method can obtain high detection rate while low false detection rate and have a high robustness for different lighting conditions and background variations.