[attachment=11831]
ABSTRACT
The detection of moving object is important in many tasks, such as video surveillance and moving object tracking. Although there are some methods for the moving object detection, it is still a challenging area. In this paper, a new method which combines the Kirsch operator with the Optical Flow method (KOF) is proposed. On the one hand, the Kirsch operator is used to compute the contour of the objects in the video. On the other hand, the Optical Flow method is adopted to establish the motion vector field for the video sequence. Then the Otsu method is implemented after the Optical Flow method in order to distinguish the moving object and the background clearly. Finally the contour information fuses the information of motion vector field to label the moving objects in the video sequences. The experiment results prove that the proposed method is effective for the moving objects detection.
I. INTRODUCTION
Moving object detection is the first step in video analysis. It can be used in many regions such as video surveillance, traffic monitoring and people tracking Generally speaking, there are three common motion segmentation techniques, which are frame difference, background subtraction and optical flow method. Frame difference method has less computational complexity, and it is easy to implement, but generally does a poor job of extracting the complete shapes of ertain types of moving objects . Background subtraction method uses the current frame minus the reference background image. The pixels where the difference is above a threshold are classified as the moving object. The Mixture of Gaussians method is widely used for the background modeling since it was proposed by Friedman and Russell . Stauffer presented an adaptive background mixture model by a mixture of K Gaussian distributions. Optical flow method can detect the moving object even when the camera moves, but it needs more time for its computational complexity, and it is very sensitive to the noise. The motion area usually appears quite noisy in real images and optical flow estimation involves only local computation . So the optical flow method can not detect the exact contour of the moving object. From the above it is clear that there are some shortcomings in the traditional moving object detection methods:
• Frame difference can not detect the exact contour of the moving object
• Optical flow method is sensitive to the noise.
The KOF method which is proposed in this paper can solve the above problems. KOF method uses the Kirsch operator to acquire the boundaries information of the moving objects, meanwhile the optical flow method is used to get the motion vector field of the moving objects.
Then both of the information acquired above is fused. At last, the moving objects are labeled with the minimum rectangle outside. The experiment results show that the present method is effective.
2. PROPOSED MOVING OBJECT METHOD
A. The outline of the method
The process of KOF method is shown in Fig. 1. The proposed method mainly consists of the edge detection, optical flow, data fusion and morphologic operation. Consider the requirements of the simplicity and
effectiveness, Kirsch operator is used for the edgedetection. For the task of the optical flow, the Lucas- Kanade method is adopted, which can quickly
provide the dense optical flow vector of the moving object.
The binary process adopts the Otsu algorithm . It can decide the threshold which is used to distinguish the background and the moving objects self-adaptively. However, because of the noise, the optical flow method can not detect the accurate boundaries of the moving objects. The edge detection algorithm mentioned just before can solve this problem. Moreover, the edge image acquired by the Kirsch operator can be regarded as space gradient, while the optical flow image is time gradient . Combining the space gradient information with time gradient information can give us the more accurately
information of the moving objects, so in the data fusion, the AND operator is used between the edge binary image and the optical flow binary image. In order to get the more exact contour of the moving objects, the morphologic
operations such as Close and Hole Filling are implemented. Finally, the moving object is extracted from the image.
B. The edge detection method
Kirsch operator
The Kirsch-operator is a non-linear edge detector that finds the maximum edge strength in a few predetermined directions. Mathematical description
The operator is calculated as follows for directions with 45° difference:
where the direction kernels
and so on.
The edge image can be regarded as the space gradient. There are some gradient operators, such as Sobel, Robert, Kirsch etc. As Kirsch operator can adjust the threshold automatically according to the character of the image, the Kirsch gradient operator is chosen to extract the contour of the object. The Kirsch operator has eight window templates. Every template makes the greatest response to a particular direction. The eight template operators are shown in Fig. 2. Except the outermost column and the outermost row, every pixel and its 3×3 eight neighborhoods in an image convolved with these eight templates respectively, so every pixel has eight outputs, the maximum output of the eight templates is chosen to be the value in this position. The gray value of a point and its eight neighborhoods in the image are illustrated