Integrate Interval Frame Difference Method

TELKOMNIKA ISSN: 1693-6930  An Improved Gaussian Mixture Model Method for Moving Object Detection Yujian Wang 119 the upper left corner is selected to be processed. If it is still unsuccessful, the match is carried out in accordance with the current pixel [9]. Algorithm processing steps are as follows: Step 1, judge whether the previous pixel belongs to the background region, otherwise go to step5. Step 2, select a 3  3 pixel block with the pixel in the middle of the block to calculate its average value 3 t x , and retain the average value 2 t x of 2 x 2 pixel block. Step 3, match 3 t x with its Gaussian distributions. If the match is successful, the parameters of Gaussian distributions are updated, then go to the next pending pixel, otherwise go to step4. Step 4, match 2 t x with its Gaussian distributions. If the match is successful, the parameters of Gaussian distributions are updated, then go to the next pending pixel, otherwise go to step5. Step 5, use the current pixel for processing, and then go to the next pending pixel.

4.2. Integrate Interval Frame Difference Method

In the detection of video moving target, the inter-frame difference method calculates the difference between two frames of the adjacent frames to obtain the range of motion caused by the moving object. The interval frame difference method is used to replace the common frame difference method, and the time interval between two frames is increased [13]. In a certain extent, the shortcomings of the common frame difference method can be overcome, and the noise of the region is eliminated effectively. Because the inter-frame difference method only retains the relative information, for the slow-moving objects, it is hard to detect the overlapping portions of the two frames, and it is also prone to the empty phenomenon. In order to solve this problem, this paper does an operation of 4 neighborhoods of the pixel block based on the interval frame difference method [14, 15]. Select two frame images in the video sequence. , y x f t is current frame image, , 2 y x f t  is the two previous frame image. It can detect the change of the region between two frames by the interval frame difference method. The equation is shown as follows. | , , | , 2 y x f y x f y x D t t t    12 The operation of the 4 neighborhood of the pixel block in the change region of two frames is carried out. The table is shown as follows. Table 1. Four neighborhood of , y x f 1 ,  j i x t , 1 j i x t  , j i x t , 1 j i x t  1 ,  j i x t The sum of four neighborhood of , y x f is calculated according to the following equation. , 1 1 , 1 , , 1 , y x D y x D y x D y x D y x t t t t t          13 The detected change region Ad of current frame can be divided into the coverage region Ac of moving object at the time t and coverage region background exposure region Ac at the time t-2, then the region not changed is the background region Ab. The change region and background region are classified as following equation.  ISSN: 1693-6930 TELKOMNIKA Vol. 14, No. 3A, September 2016 : 115 – 123 120          otherwise Ab T y x T y x D Ad j i x t t t , , , , , 3 2 14        M n x M n y t n n M M y x D T 1 1 2 , 15 Where , j i x t is the pixel block of video frame at the time t, T2 is the threshold of detection and T3 is the threshold of four neighborhoods of difference image. In the case of noise and interference are uniform distributions, the SNR signal to noise ratioof the video frame is large, or the value of , y x D t is large, then the value of T2 is also large, on the contrary the value of T2 is small. If 2 , T y x D t  , the pixel block can be judged as the background region, otherwise it can be judged as the change region between two frames. After the above-mentioned distinction, what needs to be done is matching the pixel block in the region of Ad with former B Gaussian distributions which are sorted in descending order according to priority.              As y x x y x x Ac y x x y x x t t i t i t t t i t i t , , 5 . 2 | , | , , 5 . 2 | , | 1 , 1 , 1 , 1 ,     16 Where 0iB, 1 ,  t i  is the mean of the ith Gaussian distribution block at time t-1, 1 ,  t i  is the variance. If it satisfies the matching condition, it belongs to the background exposure region Ac, otherwise it belongs to the coverage region As. The improved interval frame difference method is integrated into Gaussian mixture model. The complete region of the moving target is extracted by using 4 neighborhood of the pixel block of the traditional frame difference method. It effectively solved the problem of the traditional Gaussian mixture model that cannot complete and accurate detection of large and slow moving targets. In the early stage of the establishment of the Gaussian mixture model, it can greatly reduce the amount of computation in the process by introducing block thought. However, the phenomenon of void, non connection etc is solved by improved interval frame difference method.

4.3. Selection of Background Update Rate