Analysis of Distracted Driving Detection Based on Deep Learning Human Posture.

Distracted driving is a phenomenon that drivers' attention points to activities unrelated to normal driving, which leads to the decline of driving operation ability. With the popularity of smart phones, traffic accidents caused by distracted driving are on the rise. Traffic accidents caused by driver's inattention due to vision deviation or distraction are particularly obvious in rear end collisions. The results of the distracted driving questionnaire show that almost all drivers have experienced distracted driving. In order to detect whether drivers are distracted when driving, this paper proposes a distracted driving analysis based on deep learning of human posture. This method uses the MMDetection framework in depth learning to build a target detection network to achieve target detection, then uses the depth neural network to estimate the geometric transformation features, and uses the stacked hourglass network to accurately predict the state features of drivers in the graph. The experimental results show that the measurement method has good accuracy, fast speed, good robustness and consistency, more accurate extraction of human posture related features, strong generalization ability and certain practical value.