Detecting Preceding Vehicles Using 4-Dimensional Mapping of Colors in Image

Vision-based vehicle detection has received increasing attention in recent years in the framework of advanced driver assistance systems. However, the variability of vehicle and background poses an enormous challenge. In this paper, an approach used 4-dimensional mapping of RGB colors and integrated with corners and edges features is proposed to detect preceding vehicles in images, addressing the shortage of existent vision-based methods in the environment with complicated background and different luminance. Firstly RGB colors in images are mapped into a 4-dimensional space and therefore the corresponding positions subjected to vehicles or background can be classified by support vector machine, of which parameters are optimized with Particle Swarm Optimization algorithm. Hence, by using morphological processing, hypothetical areas of vehicles can be preliminary segmented. In addition, corners and edges features in RGB images are useful to verify vehicle hypotheses, so the feature matrixes integrated with corners and edges of hypothetical areas are calculated as to obtain feature vectors after reducing dimensions, the feature vectors from different hypotheses can be classified and thus the very vehicles areas can be labeled. Experiment results show that this novel approach has a better performance in complicated backgrounds and enervates the adverse effect of illumination with different intensity at the same time. It achieves an average accuracy of 94.1%.