Fusion of Hand Landmarks Detection and Body Region via Semantic Segmentation into Visual Cue for Hands Tracking over Steering Wheel.

Aiming at hands tracking over steering wheel, fusion of hand landmarks detection and body region segmentation into visual cue has been proposed in order to make likelihood evaluation simpler and efficient. Labelled multi-Bernoulli filter in Track-Before-Detect manner utilizes the visual cue by separable likelihood to evaluate pixels categorized into hands and body region. Experimental result has been shown to demonstrate performance of the proposed method.