Moving object detection using a moving camera has long been a highly challenging task in computer vision. In this paper, we propose a different method for detecting a moving object by means of the normal flow. The normal flow vectors are directly calculated from two consecutive frames without any constraints. Unlike some traditional methods which usually rely on feature correspondences establishment or optical flows estimation, our proposed method does not have these constraints. Those commonly used assumptions such as smoothness and continuity are no longer needed in our algorithm also. In other words, it is not required for a captured scene which has highly textured structure and distinct features by using our proposed algorithm. Our proposed method consists of three main components: 1) an image is segmented using the mean-shift algorithm, 2) an initial labeled field is then derived by examining the normal flow vectors within each region in the segmented image, and 3) the Markov Random Field (MRF) and the graph-cut optimization are separately applied to obtain the final labeling for each image. Experimental results demonstrate that the proposed algorithm is efficient in detecting moving objects.