Detection of moving targets--research status

Transfer from Http://blog.csdn.net/tfygg/article/details/51760640?utm_source=itdadao&utm_medium=referral

Moving target detection is the method of computer vision to reduce the time and space redundancy information in video, and effectively extract the process of the object with space position change. It has been a very popular research area, and a quick search with "motion detection" on the IEEE Xplore will return more than 18,000 documents. After decades of efforts of researchers, moving target detection technology has achieved good results, widely used in intelligent monitoring, multimedia applications and other fields. At present, the international research in this area of the authoritative magazine: Pami (Transaction on the Pattern Analysis & Machine Intelligence), IVC (Image and Vision Computing), but also There are some important academic meetings: CVPR (IEEE Computer Society Conference on Computer Vision and Pattern recognition), ICCV (International Conferen Ce on Computer vision), ECCV (European Conference in computer vision), IWVS (International Workshop on Vision surveillance), etc. 。 Over the years, according to the application, technical methods and other aspects of different, scholars have proposed a number of different methods of moving target detection to adapt to the complex and changeable environment.

As early as the late 70, Jain and others in the literature [9] proposed the use of inter-frame difference method to extract moving objects. The frame difference method has strong robustness to the scene containing the moving object, and the operation speed is fast, but the method can not detect all the pixel points of moving object completely, and often appears "void" phenomenon inside the detected moving object, so this method is suitable for the simple motion detection situation.

        for the shortcomings of the inter-frame difference method, the authors propose a concept of moving object detection based on statistics theory, namely: firstly, a background model is established by the statistics, and then the background difference method is used to classify the pixel points of moving target and background. Gloyer and other people in the literature [10] adopted this idea, the background model using the median method, that is, the use of continuous n-frame image sequence of the median value of the pixel as a background model. However, this method consumes a large amount of memory, the computational capacity is not small, and in the environment of light changes are often biased. In view of this limitation, 1997 Wren and others proposed using the Gaussian background modeling method [11], using the threshold value to determine whether the pixel is a foreground. However, backgrounds are often complex multimode situations (such as swaying leaves, etc.), so using Gaussian models often does not accurately describe the background model. Therefore, in 1999, Stauffer and other people proposed the classical mixed Gaussian background modeling method [12], this method usually can adapt well to the complex scene, and through the model parameter automatic calculation to adjust the background model, but because increases the Gaussian distribution number, the computation quantity also increases. In the past more than 10 years, a lot of improved algorithm based on mixed Gaussian model, such as: 2004 Zivkovic and other people in the literature [13] proposed the number of Gaussian model adaptive algorithm, so that the algorithm efficiency, robustness can be improved. However, the background of the actual situation is often rapid change, sometimes does not conform to the Gaussian distribution, so using the Gaussian model for background modeling will create problems. Therefore, in 2000, Elgammal and other people proposed a non-parametric motion detection method based on kernel density estimation [14], which does not need to make any assumptions about the density distribution of the background, but uses the standard kernel function to accurately estimate the pixel point and extract the moving target by the recent image sample information. The experimental results show that this method has good adaptability in the complex outdoor scene, but the disadvantage is that the computational quantity is large and the real time is not good.

In 2005, Wang and others proposed a background modeling approach based on sample consistency (SACON) [21,22]. The method also avoids any assumptions about the probability estimation of the pixel values of the video sequence, but first calculates the distance between the new pixel and the sample of the background model, then counts the number of samples with similar distances, and finally determines whether the foreground is based on the number of samples. In 2009, Barnich and other people in the literature [19] also proposed a novel pixel-based moving object detection method, and named the Visual background extraction (VIBE), the algorithm directly to each pixel in accordance with a certain number of rules randomly selected a certain amount of pixel value for background modeling, Then the Euclidean distance is used to classify the foreground and background of the pixel points. The advantage of the algorithm is that it does not need to assume any probabilistic model, and it can detect moving objects in the second frame of the video sequence, and the computation speed is also very fast. In the 2011, Barnich and other people in the literature [20] in the classical motion object detection algorithm made a summary and compared with vibe, finally, through the experiment proved the characteristics of vibe high efficiency. Although this article is dedicated to a good test results, but in some dark background, shadows and fast background changes in the scene will still appear some problems, such as "ghost" phenomenon. In 2012, Hofmann and others in the literature [23] first proposed PBAs (pixel-based Adaptive segmenter) Moving target detection method. According to the advantages of Sacon and vibe two algorithms, the algorithm has improved the accuracy of target detection, but the calculation of the algorithm is large and the real-time is not good.

From the above introduction, we can see that the research of moving object detection method based on statistics theory has undergone a process from simple model to complex model to simple model. In the rapid development of this kind of moving target detection method, the scholars also put forward many kinds of moving target detection methods, such as the method based on the clustering theory, the method based on the fuzzy theory, the background prediction method, the neural network-based method and the optical flow method, etc. [15-18,38,39].

The classical algorithm of moving object detection based on clustering is the Codebook method presented by Kim and others in the literature [15] in 2005. Compared with the MOG,KDE algorithm in that period, Codebook does not use probabilistic model, but uses codebook to classify pixels and achieve the goal of extracting foreground. This method can also adapt to a certain complex scene. However, due to the complexity of the scene, the code word will continue to increase, which will lead to excessive memory consumption, real-time is also subject to certain limitations.

In recent years, some scholars have proposed using fuzzy theory to solve the inaccuracy and uncertainty of background subtraction in view of many uncertain factors in actual scenes. According to the different fuzzy theory, this kind of method is divided into traditional fuzzy background modeling and two type fuzzy background modeling [38,39]. The experimental results show that the method of background modeling based on fuzzy model has good robustness in complex scenes such as illumination change and dynamic background, but its disadvantage is that the increase of computational amount will also consume more memory.

Background prediction refers to the use of filters to estimate the background, which is considered a foreground if a pixel value of the current frame deviates from its predicted value. The reference [27] uses the Kalman filter to estimate the background value, the method can adapt to the situation where the light changes rapidly, but the detection accuracy is not high, and when the moving target movement speed is slow, the detection result will often appear "drag shadow" phenomenon.

2008, Maddalena and others in the literature [24] a background subtraction method (sobs) based on self-organizing neural network was proposed. The algorithm generates a neural network background model by self-organizing method, and then extracts the moving target by the distance between the pixel points of the current frame and the background model. Then, Maddalena and other people to improve the sobs, put forward the sc-sobs[25], the spatial consistency into the background update stage, and further improve the robustness of the algorithm.

Before 2006, scholars put forward many methods of moving object detection based on "pixel" feature, few people proposed moving object detection method characterized by "region" or "frame". The texture feature is a very easy to distinguish the regional characteristics of the image, Heikkila and so on for the first time a texture histogram based on LBP for background modeling method [8], but because the texture of high computational complexity, so this kind of method is not good real-time. And the domestic research on this direction has achieved good results. 2010 Chinese Academy of Sciences Automation obtained Dr. Liao in the literature [33], a new texture description method SILTP, combined with the pattern kernel density estimation method for foreground and background segmentation is proposed. The algorithm can deal with the detection of moving objects in complex environment. and the frame-based moving object detection method uses the idea of direct video frame background modeling, and the classical algorithm has the background subtraction method [32]. The algorithm uses principal component analysis (PCA) to decompose the continuous multi-frame video and then extract the foreground, and the proposed method also opens up a new development direction of moving target detection.

There are many kinds of moving target detection algorithms, but none of them are suitable for moving target detection in all cases. Therefore, the key of moving target detection is how to find the suitable method according to the existing theory and the characteristics of the actual scene, so as to meet the needs of practical application.

"References"

[9] Jain R, Nagel H H. On the accumulative difference pictures from image sequences of real world scenes. [J]. IEEE Transactions on Pattern analysis & Machine Intelligence, 1979, 1 (2): 206-214.

[j] Gloyer B, Aghajan H K, Siu K Y, et al video-based freeway-monitoring system using recursive vehicle tracking[j]. Proc Spie, 1995, 2421:173-180.

[One] Wren C R, Azarbayejani A, Darrell T, et al pfinder:real-time tracking of the human body[j]. IEEE Transactions on Pattern analysis & Machine Intelligence, 1997, 19 (7): 780-785.

[Stauffer], Chris, Grimson, w.e.l Adaptive Background Mixture Models for real-time tracking[j]. Proc CVPR, 1999, 2:2246.

[] Zivkovic Z. Improved Adaptive Gaussian Mixture Model for Background subtraction[c]//Pattern recognition, internation Al Conference on. IEEE Computer Society, 2004:28-31.

Elgammal A, Harwood D, Davis L. non-parametric model for background substraction[j]. Proceedings of the computer Vision, 2000:751--767.

[j] Kim K, Chalidabhongse T H, Harwood D, et al. Background Modeling and subtraction by Codebook construction[c]//Image Processing, 2004. Icip ' 04. 2004 International Conference on. IEEE, 2004:3061-3064 Vol. 5.
[+] Kim K, Chalidabhongse T H, Harwood D, et al real-time foreground–background segmentation using Codebook model[j]. Real-time Imaging, 2005, 11 (3): 172-185.

[+] Li Y, Chen F, Xu W, et al gaussian-based Codebook Model for Video Background Subtraction[j]. Lecture Notes in Computer Science, 2006, 4222:762-765.

Geng L, Xiao Z T. Real Time Foreground-background Segmentation Using two-layer Codebook model[c]//Control, Automatio N and Systems Engineering (case), International Conference on. IEEE, 2011:1-5.

Barnich O, Vanogenbroeck M. vibe:a powerful random technique to estimate the background in video sequences[c]//IEEE International Conference on Acoustics, Speech & Signal processing. 2009:945-948.

[Olivier] B, Marc v. VIBE:A Universal background subtraction algorithm for video sequences. [J]. IEEE Transactions on Image processing A Publication of the IEEE Signal Processing Society, 2011, 20 (6): 1709-1724.

Wang H, Suter D. sacon:a Consensus Based Model for Background Subtraction[j]. Schools & Disciplines, 2005.

Wang H, Suter D. Background subtraction Based on a robust Consensus method[c]//Pattern recognition, International Co Nference on. IEEE Computer Society, 2006:223-226.

[Hofmann M, Tiefenbacher P, Rigoll G. Background segmentation with feedback:the pixel-based Adaptive segmenter[c]//C Omputer Vision and Pattern recognition workshops (CVPRW), IEEE Computer Society Conference on. IEEE, 2012:38-43.

Maddalena L, Petrosino A. A self-organizing approach to background subtraction for visual surveillance applications. [J]. IEEE Transactions on Image processing A Publication of the IEEE Signal Processing Society, 2008, 17 (7): 1168-1177.

[+] Maddalena L, Petrosino A. The sobs algorithm:what is the limits? [c]//computer Vision and Pattern recognition workshops (CVPRW), IEEE Computer Society Conference on. IEEE, 2012:21-26.

[+] Ebner m. Evolving color constancy[j]. Pattern recognition Letters, 2006, 27 (11): 1220-1229.

[Karmann] K. Achim von Brandt. Moving object recognition Using an Adaptive Background memory[c]//time-varying Image processing & Moving Object Recog Nition. 1990.

[Toyama K], Krumm J, Brumitt B, et al Wallflower:principles and practice of background maintenance[c]//IEEE Internat ional Conference on Computer Vision. IEEE, 1999:255-261 vol.1.

[Bouwmans T, Baf F E, Vachon B. Background Modeling using Mixture of Gaussians for Foreground Detection:a Survey[j]. Recent patents on computer Science, 2008, 1 (3): 219-237.

[Benezeth] Y, Jodoin P M, Emile B, et al comparative study of background subtraction algorithms[j]. Journal of Electronic Imaging, 2010, 19 (3): 033003-033003.

[To] T. Bouwmans, F. El-baf, and B. Vachon. Statistical background modeling for foreground DETECTION:A survey. Handbook of Pattern Recognition and computer Vision, World Scientific Publishing, 4 (2): 181–199, January 2010.

Oliver N, Rosario B, Pentland A. A Bayesian computer Vision System for Modeling Human Interactions[j]. Pattern analysis & Machine Intelligence IEEE transactions on, 2000, 22 (8): 831-843.

[Liao S], Zhao G, Kellokumpu V, et al Modeling pixel process with scale invariant local patterns for background Subtra Ction in complex scenes[c]//computer Vision and Pattern recognition (CVPR), IEEE Conference on. IEEE, 2010:1301-1306.