Driver’s Posture Detection Based on Centroid Coordinates of Two-hand(arm) Region

At present, academic research focuses on detecting driver fatigue and distraction through driver's eyes and head. Few studies detect driver's posture and behaviour through the head, hands and torso, and most of them use skin color detection to extract the corresponding area. The traditional method uses full-image pixels as features to train the neural network classifier and it has a long training time. This paper proposes a method to identify the driver's posture by extracting the centroid coordinate of the driver's two-hand (arm) region from the image, considers the situation of different regions connected and selects the KNN and decision tree classifier which have a better performance. The method can correctly recognize the driver's normal driving posture, operating the gear and the wrong posture such as one-handed driving, making a phone call and eating. This method greatly reduces the dimension of the feature data. In the case of ensuring the accuracy of recognition, the training time of the new method is shortened by 1/60000 compared with the conventional method.