Real-time Object Detection Using Distance Transforms

This paper presents an e cient shape-based object detection method using Distance Transforms (DTs). The proposed method extends previous DT-based matching techniques by using multiple features and a template hierarchy associated with a coarse-tone search over the template transformation parameters. Signi cant speed-up factors are typically obtained when comparing the proposed hierarchical method with an equivalent brute-force technique; we have measured speed-up gains in the order of two magnitudes. This brings a number of template matching applications which previously required special-purpose correlation hardware onto the realm of the ubiquitous PC. We present results on real-time tra c sign detection to illustrate our approach.