Abstract: This paper proposes a novel local feature-based object tracking algorithm. The object is segmented into a few regions using multiple scale segmentation, and each region is denoted as a blob which combines the local color, shape and location. An object appearance model is built based on the blob feature. The object is tracked by the blob matching criterion defined. The appearance model includes the local color and spatial structures, so the object can be tracked in the condition of partial occlusion and change of the object scale. Examples demonstrate that the proposed algorithm can obtain the accurate tracking results in the complex environment and has better performance than the well-known mean shift algorithm.

Key words: object tracking, appearance model, blob matching, image segmentation