Performance improvement of human detection using thermal imaging cameras based on mahalanobis distance and edge orientation histogram

Abstract

In the thermal imaging, human object detection is difficult when the temperatures of surrounding objects are similar to or higher than human's. In this paper, we propose a novel algorithm suitable to those environments. The proposed method first compute a mean and variance of each pixel value from the initial several frames, assuming that there is no object in those frames. Then for each frame after the initial frames, the Mahalanobis distance is computed between the mean value and the current frame at each pixel, and the region of interest (ROI) is estimated. Finally, using the aspect ratio and edge orientation histogram, we determine if the ROI is human or not. The experimental results show that the proposed method is effective in both summer and autumn.