A fast region-based active contour for non-rigid object tracking and its shape retrieval

Hiren Mewada, Jawad F. Al‐Asad, Amit Patel, Jitendra Chaudhari, Keyur Mahant, Alpesh Vala

Conventional tracking approaches track objects using a rectangle bounding box. Gait, gesture and many medical analyses require non-rigid shape extraction. A non-rigid object tracking is more difficult because it needs more accurate object shape and background separation in contrast to rigid bounding boxes. Active contour plays a vital role in the retrieval of image shape. However, the large computation time involved in contour tracing makes its use challenging in video processing. This paper proposes a new formation of the region-based active contour model (ACM) using a mean-shift tracker for video object tracking and its shape retrieval. The removal of re-initialization and fast deformation of the contour is proposed to retrieve the shape of the desired object. A contour model is further modified using a mean-shift tracker to track and retrieve shape simultaneously. The experimental results and their comparative analysis concludes that the proposed contour-based tracking succeed to track and retrieve the shape of the object with 71.86% accuracy. The contour-based mean-shift tracker resolves the scale-orientation selection problem in non-rigid object tracking, and resolves the weakness of the erroneous localization of the object in the frame by the tracker.