Real-time motion estimation of car driver's hands and arms' direction in vision under possible mutual occlusion by particle filter

摘要

Motion of car driver's hands and arms have been tracked in vision by an elaborated particle filter with scenes being to have possible mutual occlusion by arms. Grasping angular positions over car steering by the two hands of car driver and arm's direction from each hand are estimated in time by particle filter. The mutual occlusion may occur due to arm's position to hide the other hand/arm. The scene has been formulated in a state space model with state vector consisting of the grasping angular positions and arms' directions. Smooth motion of the state vector has been described in a system model of the state space model. Observation model evaluates likelihood of hypothesized rectangle regions of hands around the steering and arms with elimination of other hand from the captured image. Real-time implementation of the proposed method to PC with C language and OpenCV library has been conducted with tracking experiments for several scenes.