Real-Time Salient Closed Boundary Tracking using Perceptual Grouping and Shape Priors

In this paper, we propose a real-time method for accurate salient closed boundary tracking via a combination of shape constraints and perceptual grouping on edge fragments. Particularly, we encode the Gestalt law of proximity and the prior shape constraint in a novel ratio-form grouping cost. The proximity and prior constraint are depicted by the relative gap length and average distance difference along the to-be-tracked boundary with respect to its area. We build a graph using the detected edge fragments and in-between gaps. The grouping problem is formulated as searching for a special cycle in this graph with a minimum grouping cost. To reduce the search space and achieve real-time performance, we propose a set of novel techniques for efficient edge fragments splitting and filtering. We evaluate this method on a public real-world video dataset against other methods. The average alignment errors of different sequences achieved by our method are mostly less than 1 pixel, an improvement over state-of-the-art methods.