Vehicle Detection Method Based on Edge Information and Local Transform Histogram

Abstract

Robust and reliable vehicle detection from images is an important problem with applications to Advanced Driver Assistance Systems (ADAS). In this paper, we show that edge information can be used as vehicle absence cue to reject early a large portion of background windows. An improved coarse-to-fine vehicle detection method is proposed in order to achieve efficient detection with high accuracy. Furthermore, according to our study, we find that contour is the major property of a vehicle. We propose an effective method to capture the contour information, which is edge based Local Transform Histogram (LTH). We first extract the Sobel edge of the input image, and then use LTH to encode the contour information. Though linear SVM remains a popular choice for its speed and performance, we use Histogram Intersection Kernel (HIK) SVM for its better classification accuracy. Our method is evaluated on a famous benchmark dataset: UIUC car dataset. The experimental results show that our method has significantly improved the accuracy and stability of vehicle detection.