Pedestrian tracking routine for passive automotive night vision systems

Purpose - This paper aims to present a new approach to implement a pedestrian tracking algorithm for a passive automotive night vision application. Design/methodology/approach - First, the basic information of passive and active night vision systems is presented, with implementation methods adopted in previous work. The proposed thermal-image processing is a combination of seed detection, boundary detection and seed growth computations, based on a temperature thresholding scheme. Findings - The processing routine performance is assessed when implemented to a continuous sequence of thermal acquisitions, from a commercial automotive night vision module. Experimental results show good tracking performance for both pedestrians and passing vehicles. Research limitations/implications - A strategy of multi-seed growth, directional seed growth and image fusion is proposed to improve the current tracking algorithm. Originality/value - New thermal image processing routines are applied to commercial, automotive night vision modules, to provide robust pedestrian tracking at real-time processing speed.