An extraction method of moving objects by shape energy functions with prior knowledge

We propose a method for tracking moving objects and extract their contours in image sequences. In this method, we construct a new energy function by learning dynamics of shape changes, and then incorporated it into the shape energy. Moreover, we applied metric-based kernels instead of Gaussian kernels in the shape energy function. From this, the distribution of the feature vectors in the kernel space was represented parametrically, and the iterative computation cost of the shape energy function decreased. Here the B-spline curves are applied to draw the contour of the moving object. The object contours are estimated by minimizing the energy function iteratively according to the change of the energy function itself. As the result, the moving object in the image sequences was contoured and tracked smoothly. Finally, we made experiments to confirm effectiveness of this method.