Objects based change detection in a pair of gray-level images
Pattern Recognition
From moving edges to moving regions
IbPRIA'05 Proceedings of the Second Iberian conference on Pattern Recognition and Image Analysis - Volume Part I
| Hi-index | 0.00 |
We propose a motion segmentation algorithm that aims to break a scene into its most prominent moving groups. Instead of identifying point correspondences between the image frames, the idea to find what groups of pixels are transformed from one image frame to another. To do this, we treat the image sequence as a three dimensional spatiotemporal data set and construct a weighted graph by taking each pixel as a node, and connecting pixels that are in the spatiotemporal neighborhood of each other. We define a motion profile vector at each image pixel which captures the probability distribution of the image velocity at that point. By defining a distance between motion profile at two pixels, we can assign a weight on the graph edge connecting them. Using normalized cuts we find the most salient partitions of the spatiotemporal volume formed by the image sequence. Each partition, which is in the form of a spatiotemporal volume, corresponds to a group of pixels moving coherently in space and time. Normalized cuts can be computed efficiently by solving a generalized eignevalue problem.