Graph models offer high representational power and usefulstructural cues. Unfortunately, tracking objects by matching graphs overtime is in general NP-hard. Simple appearance-based trackers are able tofind temporal correspondences fast and efficient, but often fail to over-come challenging situations like occlusions, distractors and noise. Thispaper proposes an approach, where an attributed graph is used to rep-resent the structure of the target object and multiple, simple trackersin combination with structural cues replace the costly graph matching.Thus, the strengths of both methodologies are combined to overcometheir weaknesses. Experiments based on synthetic videos are used toevaluate two possible structural cues. Results show the superiority of thecue based on barycentric coordinates and the potential of the proposedtracking approach in challenging situations


Artner, N., & Kropatsch, W. (2013). Structural Cues in 2D Tracking: Edge Lengths vs. Barycentric Coordinates. In Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications (pp. 503–512). Springer. https://doi.org/10.1007/978-3-642-41827-3_63