科研成果详情

Visual tracking is a challenging task in computer vision, which intends to estimate the motion state of the target of interest in subsequent video frames. In that context, it is well-known that the rapid movement and rotation of the target affects the tracking results. This article proposes a novel single-target tracking algorithm based on the spatially weighted generalized Gaussian mixture model framework. The clustering method for color compression in preprocessing is considered to modify the frames to make them have sharper distributions. Then, the mixture models are built to express the color statistical features of the aimed area and context. The segmentation weights obtained from the responsivity of the pixels in the candidate ellipse to the target and background will guide the update of the target position and size in a heuristic way. The adjustment of models will depend on the aspect ratio change of the bounding ellipse. The performance of the proposed approach is verified on public datasets and compared with other algorithms. The experimental results show that our method achieved more accurate and robust tracking.