Introduction

This is a new automated approach that can be applied in many areas of computer vision. The goal is to allow a machine to understand what it sees when it is connected to one or more cameras. Segmentation and detection are very important problems in computer vision. We applied this method in video-surveillance, we go to track and detect multiple objects on different resolutions. However, the cases of multiple target localization, detection in non-lucid medium and tracking in multiple overlapping objects makes it diffi- cult in many computer vision domain and video analysis applications. It has critical effect in image sequence result.
By that analysis and design of mechanisms for managing real-time vision is still an open problem. In this paper, we conduct an experiment to achieve a reliable monitoring of all objects in video sequences. This paper presents efficient method for solving these problems. In this paper we use blob tracking that defines the objects detected by adding more characteristics of these objects. This algorithm has recently been adopted as an efficient technique based on blob tracking [1, 2, 3], but it does not have enough parameters for a reliable extraction of the characteristics of objects detected. The mean-shift algorithm is an efficient technique for tracking 2D blobs in an image sequence although the scale is a crucial parameter. In [2] they showed that the difference of Gaussian (DOG) average displacement of the core allows effective monitoring of blobs in the scale space. One of the drawbacks of the methods of background difference for the detection Movement is the length of processing time. Therefore, the motion detection in real time in some cases is very difficult, rather impossible with this type of method. Because Object tracking should deal with some challenges like occlusion, target appearance changes, cluttering, etc., In [3] new approach based on Blob tracking using histogram filtering, is based on meanShift theory, This algorithm needs less computational complexity compared with methods performed in the whole image. In the traditional mean-shift algorithm, the target model is unchanged during the tracking and belongs to the first frame, so this algorithm is not robust in the case of large changes of appearance, size and direction of the target. Mean-shift is a non-parametric kernel estimation, which gives good results, especially in image segmentation [4]. However, the size of the kernel must be defined beforehand. Several techniques are defined in [5, 6] and [7]. We are based on one of these methods to solve this problem of scale. We will also incorporate a notion to track-segmentation simultaneously [9] with the introduction of the theory of Graphcut [8]. That will guarantee fast in terms of performance and cost calculations weakness. In the first section we have defined all the theoretical concepts used of this solution, we recall the principles of the method Meanshift, the blob tracking, and the theory of Graphcuts. Then in the second section, we describe the algorithm proposed. And in the third section we present the experimental results of the algorithm, followed by a conclusion.

2 Architecture

The approach is typically able to react according to the content of the captured scene. In which we deal with a lot of problems namely: lighting problem and luminance change issues such as movement changes the background and changes in small claims for this we proceed by a able to make simultaneous segmentation and tracking based on storing blobs for each iteration. For this reason the system must predict the existence of the blob based on a list of blobs comparing their histograms, which is going to reduce the processing time. After this step it may have new birth of blobs. Blobs if the list is updated (correction step). It is a method to identify and trace the movements of objects; it allows determining the positions in successive frames. Before pursuing a blob, you must first identify and located. So the first step we must do is detecting blobs.
The process is performed by grouping pixels of similar brightness (contrast) and of the same color (texture). A tolerance threshold indicates the difference between the values that can appear in a blob. The difficulty is that the shape and values of blobs may change as they move. Then we should detect the blob in the following frames by establishing a significant correlation between the different blobs in each frame (Fig. 1).