A method for tracking a target object utilizing a binocular system includes capturing first and second images, the first image captured from a first camera device and the second image captured from a second camera device. A plurality of image patches are applied to the first and second images and a plurality of detection costs each associated with respective ones of the image patches applied to the first and second images is determined. A plurality of matching costs each corresponding to respective ones of selected matching pairs of image patches between the first and second images is determined. At least one cost flow path is determined from a source vertex of the first image to a sink vertex of the second image based on the detection costs and the matching costs and the target object is tracked based on the at least one cost flow path.