Dynamic Obstacle Detection of Road Scenes using Equi-Height Mosaicking Image

Today, many automobile companies and researchers have developed various safety systems to reduce fatalities by traffic accidents[1, 2]. In order to prevent traffic accidents by distracted driving, therefore, the proposed system presents the vehicle and pedestrian detection using a novel image representation called equi-height mosaicking system[3]. Furthermore, the proposed system additionally suggests the part-based side detection method using equi-height peripheral mosaicking image to detect approaching vehicles while driving. The proposed system first presents the detection method of vehicle and pedestrian using road geometry in real-time. Especially, we propose the new image representation called equi-height mosaicking image to perform the GPU-based fast vehicle and pedestrian detection. The equi-height mosaicking image is generated by using a number of equi-height images that are made by results of road geometry analysis. First of all, the proposed system performs the distortion removal. In this step, the proposed system removes the lens distortion and image skew using the distortion map in on-line. The distortion map is precomputed by distortion coefficients and skew rotation in off-line. After the distortion removal is completed, the proposed system analyzes the road scene to generate the equi-height images. The proposed system extracts the sampling position in the image. The sampling position is extracted with a regular interval based on a distance from the camera of the vehicle. Then, the proposed system estimates the height of equi-height images on a sampled position. When the road scene analysis is finished, the proposed system generates the equi-height image. The proposed system crops the image to make equi-height images using the sampled position and estimated height. Once equi-height images is generated, the proposed system resizes the equi-height images to a fixed height. Then, the proposed system concatenates the equi-height images to generate the equi-height mosaicking image. The equi-height mosaicking image is used to increase the processing speed in detection step. When the equi-height mosaicking image is generated for fast speed, the proposed system performs the GPUaccelerated 1D search based detection using HOG-based SVM classifier on equi-height mosaicking image[4]. When vehicles or pedestrians are detected, the proposed system transforms the coordinates of detected region from equi-height mosaicking image to input image. Finally, the proposed system decides the detected vehicles or pedestrians by grouping for multiple detected regions using non-maximal suppression.