Practical and advanced image processing for security and recognition by thermal distributed image features

This paper presents the research on a practical and advanced image processing for intelligent human recognition and security system by using features of thermal images to enhance safety of vulnerable pedestrian. In this system, unique characteristics of thermal images were employed to detect the pedestrian along the roadway in varied conditions. Newly distribution of temperature regions, where correlation of the center point and the center of gravity point, or the centroid of these regions were calculated subsequently in order to extract pedestrian thermal regions in images. The processes were conducted following threshold method and noise filtering of initial image. The 3-valued threshold and labeling process were applied to the initial image to divide thermal area into four main temperature regions. The experiments were conducted under four conditions; numerous pedestrian, single pedestrian and oncoming car, overlapping pedestrian and occluded pedestrian. The subjects were classified by calculating pixel ratio of thermal regions, average luminance within divided temperature regions, and correlation among the centroid of each thermal region. Results of the experiments are encouraging. In particular, the proposed intelligent recognition system utilizes feasible and low-cost approach, compared to prior works.