Search

Vehicle detection is becoming a necessary part of Automatic Cruise Control (ACC) and Advanced Driver Assistance Systems (ADAS). Our main focus in this paper is on improving the performance of single camera based vehicle detection systems. Edges are one of the main characteristics of an object, which carries most of the information about an object in an image. In this paper, it was observed that horizontal edges are strong feature for vehicle detection. Therefore, we generated initial candidate using Horizontal Edge Filtering (HEF) on canny edge map. These initial candidates are further verified using Bag-of-Features (BoF) with K nearest neighbor algorithm. A threshold is used on differences of histograms of training and test images for matching the vehicles. In this paper, the combination of edges (initial candidate) and bag-of-features (final verification) has improved detection rate significantly as compared to other well known methods. Our method has 96% detection rate on roads inside a city and 98% detection on highways.

Edge detection has been the foremost step in image processing and computer vision, because an edge representation drastically reduces the amount of data to be processed. Although classical methods of edge detection like Sobel, Canny, etc. are simple to use but has a dilemma between noise removal and edge localization. If noise is to be removed by using a low pass filter then edges are blurred. However, if edges have to be preserved then noise severly corrupts the edge map. In this paper, we have proposed a new method of edge detection, BiGaussian edge Filter, which simultaneously removes noise from real life images, while generating well localized edges. We have compared our method using images form Berkely’s segmentation data set. Experimental results show the robustness of our method to noise in real life images.