Vehicle Area Segmentation Using Grid-Based Feature Values

Abstract

We present a vehicle segmentation method for still images captured from outdoor CCD cameras. Our preprocessing process partitions the background images into a set of two-dimensional grids, and then calculates the statistical feature values of the edges in each grid. For a given vehicle image, we compare its feature values of each grid to the statistical values of the background images to finally decide whether the grid belongs to the vehicle area or not. To find the optimal rectangular grid area containing the vehicle, we use a dynamic programming technique. Based on the statistics analysis and the global search technique, our method is more systematic compared to the previous heuristic methods, and achieves high reliability against noises, shadows, illumination changes, and camera tremors. Our prototype implementation performs vehicle segmentation in average of 0.150 second, for each of 1280 × 960 vehicle images. It shows 97.03 % of successful cases from 270 images with various kinds of noises.

Prof. Kim was supported by the Korea Research Foundation Grant funded by Korean Government (MOEHRD) (R04-2004-000-10099-0). Prof. Hong was supported by the Ubiquitous Autonomic Computing and Network Project, the Ministry of Information and Communication (MIC) 21st Century Frontier R&D Program in Korea (05A3-B2-50).