High-Accuracy Joint Position Estimation and Posture Detection System for Driving

In this paper, we present a novel Advanced Driver Assistance System (ADAS) that monitors driving posture. To achieve accuracy and robustness of the system, we propose a high-accuracy method for estimating the joint positions of drivers and an algorithm for detecting abnormal driving posture. The key ideas for improving the accuracy of joint position estimation are that we train a Random Forests classifier with driver image sets that we built and we introduce specific knowledges to recognize the body parts of the driver. To achieve robustness, we invented an abnormal posture detecting algorithm that accumulates information on the most recent postures. We integrated our technologies in an actual vehicle. We evaluated the accuracy and the coverage of the abnormal posture detection of our system by conducting driving tests on a test course. The results show that this system detects four abnormal driving postures, discriminates small changes in posture, and has robustness against vehicle vibrations, environment lighting, and differences in drivers' physiques.