n-roaOd Vehicle Detection based on Hierarchical Clustering and Adaptive Vehicle Localization

Vehicle detection is considered to be a significant task in automatic driving which is regarded as a challenge and thorny issue for researchers in this field. The majority of commercial vehicle detection systems are based on radar. However, methods using radar suffer from problems such as the one encountered in zigzag motions. Image processing techniques can overcome these problems .This paper proposed an approach based on hierarchical clustering in which low-level image features are used to detect on-road vehicles. The approach introduced in this study is based on a new clustering method called teammate selection. In this clustering method, a new merging measure based on cluster center distances and gray scale values was introduced. Each vehicle was assumed to be a cluster. In traditional clustering methods, the threshold distance for each cluster was fixed; however, in the method proposed in this paper, the threshold distance is adaptive which varies according to the position of each cluster. The threshold measure was computed with bivariate normal distribution. Sampling and teammate selection for each cluster were carried out by cluster members based on weighted average. Unlike other methods which used only horizontal or vertical lines, a fully image edge detection algorithm was utilized in this study. Corner is an important video image feature which is commonly used in vehicle detection systems. However, Harris features were used in this paper to detect the corners. Furthermore, LISA data set was used to evaluate the proposed method. Several experiments were conducted to investigate the performance of proposed algorithm. Experimental results indicated good performance compared to other algorithms.