Geometric structure-constraint tracking with confident parts

Multi-level visual tracking with hierarchical tree structural constraint

“…To address this problem, Cai et al [1] model the superpixels together with their relationships as an undirected graph, and formulate the target tracking as a graph matching problem. Xie et al [19] employ a minimum spanning tree to model the geometric structure constraint between superpixels, and infer the target state by voting from confident parts of the target based on appearance similarity and structural constraint. In addition to leveraging the spatial relationships between local parts, Li et al [5] construct a relational hypergraph, which models the high-order relationships among multiple local parts across the temporal domain, and identify temporally coherent parts for the target representation.…”

“…The first term measures the probability that the superpixel c I k belongs to the target, and the second term measures the similarity between the two superpixels. Then similar to [19], we get a smoother score map D M i ðIÞ by convolving the score map D M i ðIÞ with a filter:…”

“…For general objects, although the appearance may change drastically over time due to the multiple challenges, the structure relationships inside the target remain stable [19]. So incorporating geometric constraints is very helpful to enhance the stability of a tracker.…”

“…Recently, Cai et al [1] use an undirect graph to model the relationships between superpixels. Xie et al [19] represent the structure relationships between superpixels by a minimum spanning tree. However, these two methods use only a single-layer representation, and the inter-layer structures with multi-layer representation are unexplored.…”