On GPGPU parallel implementation of hands and arms motion estimation of a car driver with depth image sensor by particle filter

Abstract

GPGPU parallel computation technology has been combined with depth image sensor such as Microsoft Xbox360 KINECT for real-time estimation of car driver's hands and arms motion with an elaborated tracking method based on particle filter. Vision observation by KINECT including depth image provides more accurate hand/arm region information, so we can extend the motion estimation method, not only on hands/wrists region with skin color cue, but also on arms region not necessarily having skin color, based on a depth signal. In addition, with particle filter for state estimation in robust and in sequentially with GPGPU parallel implementation for real-time computation, it allows us to develop a real-time motion estimation system of a car driver. Contribution of this paper is twofold; 1) to provide whole summary of steering hands / arms motion estimation methods so far based on particle filters and partially with the aid of GPGPU technology, and 2) to propose a new system implementation of GPGPU parallel particle filter not only for hands/wrists region but also for arms region of a car driver with the aid of depth image sensor. Some experimental results have been shown with the proposed implementation.

The particle filter algorithm: parallel implementations and performance analysis over Android mobile devices

“…They present a parallel implementation of the algorithm to execute over digital cameras. FPGA and GPGPU implementations of the algorithm are presented in and , respectively. In , the authors proposed a parallel implementation to tracking of leukocytes (white blood cells) .…”