Citation Context

...cular, we search for a shape template that simultaneously maximizes the Chamfer score and the overlap with the hypothesized segmentation. The overlap can be expressed by the Bhattacharyya coefficient =-=[3]-=-, which measures the affinity between two distributions. Assuming a uniform distribution for the points inside the shape template s, shifted to location q, we compare its overlap with the hypothesized...

Citation Context

...ll gray images of crowded real-world scenes. For each pedestrian, we want to detect its presence, even when it is partially occluded, and precisely localize it in the image. Following the task set by =-=[1]-=-, we also want to answer the question how many pedestrians are present in the scene. We therefore only accept a single hypothesis per pedestrian as correct and count each additional hypothesis on the ...

Citation Context

...ited notion of global consistency. In the following, we want to enforce this consistency by adding the information from global shape cues. 4.1. Chamfer Matching Chamfer matching was first proposed by =-=[2]-=- and later refined by [5, 19, 11] to detect objects based on global shape features. Given a set of trained shape templates (e.g. object silhouettes), Chamfer matching searches the image for locations ...