Real-time vehicle detection using equi-height mosaicking image
Min Woo Park, Jung Pil Park, Soon Ki Jung · 2013
In this paper, we present a real-time forward vehicle detection warning system using a novel image representation called an equi-height mosaicking image. The proposed system uses a GPU (graphic processing unit) based approach for the real-time processing of a road scene image captured from a single camera. The equi-height mosaicking image improves the execution time of the existing GPU-based acceleration approach without decreasing the detection accuracy. The equi-height image is generated as follows. After a geometric analysis of a road scene using the vanishing point and horizon, we crop a set of image strips by sampling several positions on the road at uniform intervals. The height of each image strip is computed by projecting the predefined height of a vehicle at a distant position onto an image plane. After all the cropped images are resized to the uniform height required to build the equi-height image, we concatenate these resized images, similar to a panorama image, to create the equi-height mosaicking image. The concatenated image has a long width but the height of the image is uniform. The proposed system then performs a GPU-based vehicle detection on the concatenated image using a 1D search based support vector machine (SVM) classification. The proposed method is faster than the GPU-based OpenCV HOG detector because of the reduced search area.