Accurate Segmentation of Moving Objects Using Object Contours
Sang-Myoung Ye, Rae‐Hong Park · ITC-CSCC :International Technical Conference on Circuits Systems, Computers and Communications · 2007
Active contour algorithms have been used for segmentation of moving objects. This paper proposes a segmentation method of moving objects using object contours. It focuses on improving the accuracy of segmentation performance, in which a modified image force model is employed in the active contour algorithm. The proposed image force model utilizes the color difference between the background frame and each frame of the input sequence. The proposed method starts with determining four main (coarse) directions (left, right, up, and down) at each pixel of the initial object contours that are to be shrunk inwards to the object center. Then, the sets of pixels to represent five sub (fine) directions (0, ±tan?¹0.5, and ±tan?¹1) with respect to each of the main directions are defined. As the (gray-level) almost-exact object contour map, along five sub-directions we employ the sum of vector angles in the RGB color space between the background frame and each frame of the input sequence. Finally, our result (the almost-exact object contour map) is used as a new external force model for an active contour algorithm to convert initial object contours into exact object contours. The proposed method can provide more detailed and improved object segmentation results than the conventional method. Experimental results with a number of test sequences show the effectiveness of the proposed method.