Research On Motion Target Detection Algorithm

Computer vision is a subject which is consist of multidisciplinary and multidisciplinary fields,such as aviation,medical testing,industrial robots and so on.Moving object detection is one of the most important research directions in the field of computer vision.It relates to the recognition and tracking of moving objects in the later period and is a very important part of the intelligent video monitoring system.Therefore,it is very important to extract the moving target quickly and accurately.Moving target detection algorithm aims to extract the desired goal quickly and accurately.At present,the most popular moving object detection algorithm are Gaussian Mixture Model(GMM),Visual Background Extractor(ViBe)and so on.Compared with Gaussian Mixture Model(GMM),the Vibe detection algorithm is simple and easy to implement,and it has high robustness.Therefore,it has been favored by many scholars.However,since the scene of the outside world is complicated and changeable,it is very hard to have an algorithm that can deal with all the problems in scene detection,such as light changes,camera shake,and periodic moving objects.Therefore,some scholars propose to change the way that the traditional camera acquire the image information to use the depth camera,and using the depth information to extract the moving target has less influence on the scene change.However,depth cameras are generally expensive and camera parameters are fixed,which making it difficult to adapt to large scenes or video surveillance.Based on the above problems,after reading a large number of documents,in order to improve the detection efficiency of moving targets,the traditional moving target detection algorithm is improved:Firstly,considering ghost and the slowness of the original vibe algorithm in eliminating ghost,this paper improves the original algorithm according to the properties of ghost,accelerates the elimination rate of ghost,and the problem of the moving target become still and then quickly into the background.The strategy of updating pixel sampling points using two different update rates is proposed: the update rate of the internal pixels of the stationary moving targets(suspected "ghost")and all the pixels of the normal moving targets are kept normal;and for the stationary moving targets pixels are updated with faster update rates,which speed up the ghost disappearing.Experiments show that without improving the efficiency of the original algorithm,the proposed improvements speed up the speed of original ViBe algorithm to eliminate the ghost,through experiments in different scenarios,the improved algorithm to eliminate the ghosting algorithm is 3-6 times to the original algorithm.Secondly,the problems of the immature depth camera hinder the performance of the upper algorithms.For this,a new efficient stereo matching algorithm based on random selection strategy,which is to replace the depth camera,was proposed.In this algorithm,a matching point samples set was firstly built.The samples in it were randomly selected every single frame,and their matching cost and confidence value were subsequently computed.Then,the set was updated at iterations,and confidence values was propagated around adjacent samples.In this way,the selection probability of the neighbourhoods can be changed in next iteration,and the depth map will be obtained through a coarse-to-fine pattern.Last,the depth information was added to the conventional ViBe algorithm,which led to an improved algorithm.The experimental results show that in the resolution of 400 × 300 the improved algorithm can achieve a nearly real-time frame rate of 8-10 frames per second and get the tripe speed of conventional ViBe on removing ghost and suppressing pseudo-target caused by light changing.Meanwhile,the improved algorithm obtains better detection and shadow elimination performances in common scene.In summary,the two improved algorithms quickly solve the traditional vibe algorithm to eliminate ghost and stationary object slower,and improve the efficiency of moving targets detection.And the stereo matching algorithm is used instead of the depth camera to obtain the depth information,which saves the cost effectively and avoids the influence on the detection result caused by the change of scene illumination.Achieve the goal of improving the real-time detection of moving objects and accuracy.