Detection Method Of Handheld Phone Use By Driver Based On Machine Vision

The behavior using hand-held cellphone while driving influences traffic safety, and it is one of the major causes of road accidents. Test results released by a British institute of transportation show that using a cellphone while driving, when the brain’s reaction speed slower than 30% of drunk driving, the risk of a car accident than a cellphone while driving more than four times higher than normal driving, 70% of a fatal accident was caused by the driver’s inattention. Currently, monitoring systems use the vehicle speed and mobile communications signals to determine whether mobile phone is used in the vehicle. These systems can’t distinguish between driver and passenger who use cellphone in the vehicle. Computer vision based methods can identify the behavior using hand-held cell phone by driver’s skin color analysis. Camera monitoring of a given driver’s skin color feature status has been proved to be the most promising technology due to good accuracy, real-time performance and non-intrusiveness.The Adaboost algorithm was used to detect face region in the driver’s facial image, and Lucas-Kanade algorithm was used to track the face region via the feature points selected by F-B Error algorithm in case that Adaboost algorism failed to detect face region. Adaptive color range extraction algorithm was proposed to detect the skin-color pixels in the region of interest determined according to the face region, and the percentage of skin-color pixels in the region of interest was used to estimate whether the hand is in the region. Finally, the handheld phone using behavior was identified by the duration of the state that the hand is in the region of interest.According to deep analysis about the impact of skin color in the actual driving environment conditions, the driver’s face picture in the current frame was used to create a dynamic skin color model, and then the whole skin color of driver was extracted based on the model. In order to improve the recognition accuracy, the static skin color model was used to extract the driver’s skin color for the circumjacent pixels around the pixels identified by dynamic skin color model. During the experiment found that Adaboost face detection algorithm is not very satisfactory, in order to improve the driver’s face detection rate, the time of this face detection failed screened start tracking and positioning using F-B Error algorithm.Based on the assumption that the features extracted from sequential images detect behavior of driver using cellphone while driving, this paper presents three-mode discrimination logic. Experiments show that the result using of three-mode discrimination logic greatly reduce misjudgment behavior.