Sand grains tracking in dense grain image sequences

In experiments of soil mechanics, it is important to study the motion of sand grains at high pressures. Technologies of image segmentation and object tracking can help to resolve this problem. Aiming at the dense sand grains images, this paper proposes a method to track sand grains from image sequences. A novel approach of dense grain segmentation based on boundary exploration is proposed. Firstly, adaptive Canny is employed to detect edges. Mathematical Morphologic methods are utilized to eliminate the noises. Secondly, seeds are chosen based on the histogram in each sub-image. And then rays are emitted from each seed to explore the boundary. The false boundary points are identified using an estimation strategy and modified automatically. After the contours of grains are achieved, the detected sand regions are tracked based on Multi-Feature Multi-Layer matching. Several experiments are performed. Promising segmentation and tracking results can be obtained on dense grain images.