Hand activity is a critical monitoring component in understanding a driver's behavior within the car. Current vision-based hand detection algorithms perform poorly in naturalistic settings, due to various challenges such as global illumination changes and constant hand deformation and occlusion. To achieve a more accurate and robust hand detection system, this paper presents a hierarchical context-aware hand detection algorithm, which explicitly explores context cues in the vehicle such as prevalent hand shapes and locations, preferred driving habit and coupling effect between multiple hands. The proposed context-aware hand detection algorithm significantly outperforms the state-of-the-art on the VIVA hand dataset.
Hierarchical context-aware hand detection algorithm for naturalistic driving
2016-11-01
1236329 byte
Conference paper
Electronic Resource
English