Fast and compact driver pose estimation

Abstract

Driver pose estimation is a key component in driver monitoring systems, which is helpful for driver anomaly detection. Compared with traditional human pose estimation, driver pose estimation is required to be fast and compact for embedded systems. We propose fast and compact driver pose estimation that is composed of ShuffleNet V2 and integral regression. ShuffleNet V2 can reduce computational expense, and integral regression reduce quantization error of heat maps. If a driver suddenly gets seriously ill, the head of the driver is out of view. Therefore, in addition to localizing body parts, classifying whether each body part is out of view is also crucial for driver anomaly detection. We also propose a novel model which can localize and detect each body part of the driver at once. Extensive experi-ments have been conducted on a driver pose estimation dataset recorded with near infrared camera which can capture a driver at night. Our method achieves large improvement compared to the state-of-the-art human pose estimation methods with limited computation resources. Futhermore, We perform an ablation study of our method which composed of ShuffleNet V2, integral regression, and driver body parts detection. Finally, we show experimental results of each driver action for driver monitoring systems.