Vehicle detection and localization in infra-red images

Abstract

This paper presents an algorithm for detecting vehicles in FIX images. Initially the attention is focused on portions of the image that contains hot objects only. These areas are then selected and refined using aspect ratio and size constraints about vehicles; even situations with overlapping vehicles are considered. The result Is further investigated exploiting specific vehicle thermal characteristics. A simple tracking phase is performed to improve the detection results. Thanks to the knowledge of camera intrinsic parameters the distance of vehicles is computed using an assumption about vehicles width. The system proved to be effective in different scenarios, but further tests are required to validate it in a wider range of weather conditions. It is able detect vehicles in front of the vision system in the range 25 m-100 m at a 12 Hz processing rate.