A Hybrid Object Tracking Method for Scaled and Oriented Objects

In this article, we demonstrate a hybrid tracking method for scaled and orientated objects in video. Histogram-based mean shift can track objects quickly. However, it does not utilize shape or dynamics information and typically employs a convex bounding box, such as a rectangle or an ellipse that is decoupled from object scale and orientation. Active shape models and particle filters cope with issues that emerge in the tracking of deformable objects or those whose projected silhouette changes over time. However, they require high computational complexity especially in case of increasing of numbers of landmarks or particles. Combining the strength of mean shift using nonconvex bounding boxes, object dynamics estimation with particle filters, and aligning shape information in consecutive frames using an active shape model, we can track scaled, rotated, and translated objects accurately and efficiently. Experimental results show that the proposed method can realize consistent object tracking in realistic scenarios.