Tracking of moving objects based on graph edges similarity

This paper suggests a novel contour-based algorithm for tracking moving objects in a video sequence. The algorithm uses the segmentation results of the input frames, which are represented by two region adjacency graphs (RAG) data structures. Based on the image segmentation result, the object's contour is divided into subcurves while junctions of the contour are derived. The junctions are being matched in a search area that is the RAG edges of the consecutive frame. Each pair of matched junctions may be connected by several of paths (edges) that are candidates to represent the tracked contour. Using algorithm for finding the k-shortest paths between two nodes these paths are obtained. Then, the construction of the tracked contour is done by a matching process between edges, which are the subcurves, and sets of candidate paths. The use of RAG to construct the tracked contour enables an accurate representation of the moving object while preserving efficient complexity.