Real-time stereo-based head detection using size, shape and disparity constraints

Abstract

A real-time stereo-based head detection algorithm is presented and evaluated as an occupant posture analysis system for "smart" airbag deployment. The algorithm uses several constraints to limit the number of head ellipse "candidates" found in the image. These constraints are based on shape, size and depth of the occupant's head. Results of ground truth experiments show that the head detection can accurately estimate the three-dimensional location of the occupant head. Extended experiments illustrate the robustness of the algorithm to poor lighting conditions, to occlusion, and to the presence of other competing head-like objects in the scene.