Enhancement Of Hough Voting By Using Appearance Similarity For Object Detection

In this paper, we propose an effective method for the detection of objects, such as vehicles or pedestrians, in static images. Hough Forest based object detection methods have received a lot of attention in recent years, and have achieved the state-of-the-art detection accuracy. In this study, rather than treating each voting element in the leaf equally as most of the previous works have done, we improve the voting scheme by distributing each voting element an additional weight based on the similarity between the patch and the voting element. In this way, the votes to the object of interest can be enhanced and the votes to the background area can be reduced. We demonstrate that our method can achieve the state-of-the-art performance.