An Efficient Neural Network Based Background Subtraction Method
01 Jan 2013-pp 453-460
TL;DR: The paper presents a neural network based segmentation method which can extract moving objects in video using a multilayer architecture so as to match the complexity of the frames in a video stream and deal with the problems of segmentation.
Abstract: The paper presents a neural network based segmentation method which can extract moving objects in video. This proposed neural network architecture is multilayer so as to match the complexity of the frames in a video stream and deal with the problems of segmentation. The neural network combines inputs that exploit spatio-temporal correlation among pixels. Each of these unit themselves produce imperfect results, but the neural network learns to combine their results for better overall segmentation, even though it is trained with noisy results from a simpler method. The proposed algorithm converges from an initial stage where all the pixels are considered to be part of the background to a stage where only the appropriate pixels are classified as background. Results are shown to demonstrate the efficacy of the method compared to a more memory intensive MoG method.
Citations
More filters
TL;DR: The recently developed idea of representative learning is elegantly employed to extend the existing single-layer self-organizing map background model to a multilayer one (namely, the proposed SMSOM-BM), which gains several merits including strong representative ability to learn background model of challenging scenarios, and automatic determination for most network parameters.
Abstract: In this paper, a new background modeling method called stacked multilayer self-organizing map background model (SMSOM-BM) is proposed, which presents several merits such as strong representative ability for complex scenarios, easy to use, and so on. In order to enhance the representative ability of the background model and make the parameters learned automatically, the recently developed idea of representative learning (or deep learning) is elegantly employed to extend the existing single-layer self-organizing map background model to a multilayer one (namely, the proposed SMSOM-BM). As a consequence, the SMSOM-BM gains several merits including strong representative ability to learn background model of challenging scenarios, and automatic determination for most network parameters. More specifically, every pixel is modeled by a SMSOM, and spatial consistency is considered at each layer. By introducing a novel over-layer filtering process, we can train the background model layer by layer in an efficient manner. Furthermore, for real-time performance consideration, we have implemented the proposed method using NVIDIA CUDA platform. Comparative experimental results show superior performance of the proposed approach.
37 citations
Additional excerpts
...(Corresponding author: Xuebo Zhang.)...
[...]
25 Dec 2015
TL;DR: A comparison between the existing background subtraction methods has been employed with some factors like computational time or memory requirements, and a view of the strengths and drawbacks of the widely used methods is provided.
Abstract: Background subtraction is one of the most important parts in image and video processing field. There are some unnecessary parts during the image or video processing, and should be removed, because they lead to more execution time or required memory. Several subtraction methods have been presented for the time being, but find the best-suited method is an issue, which this study is going to address. This paper presents a comparative study of several existing background subtraction methods which have been investigated from simple background subtraction to more complex statistical techniques. The goal of this study is to provide a view of the strengths and drawbacks of the widely used methods. The methods are compared based on their memory requirement, the computational time and their robustness of different videos. Finally, a comparison between the existing methods has been employed with some factors like computational time or memory requirements.
7 citations
01 Oct 2017
TL;DR: The problem of background subtraction is addressed using a single static camera using a temporal averaging of individual pixels over a small training sample and the modeling of pixel intensities with a log-normal probability density function that best fits the divergence among background pixels.
Abstract: In this research, the problem of background subtraction is addressed using a single static camera. Aside from the practicality of distinguishing foreground moving objects from background scenes, background subtraction is an essential step towards classifying and tracking objects in complex and dynamic environments. Our proposed method is based on the temporal averaging of individual pixels over a small training sample and the modeling of pixel intensities with a log-normal probability density function that best fits the divergence among background pixels. Our method has been tested in a series of different and challenging environments with illumination changes as well as high speed foreground objects with the view to be used in autonomous vehicles applications for pedestrian and car detection. The results from this research are juxtaposed against the state-of-the-art methods and demonstrate the efficiency of our approach.
7 citations
Cites background from "An Efficient Neural Network Based B..."
...A multilayer neural network approach also appears in [20] where a neural network combines the spatio-temporal correlation among pixels....
[...]
TL;DR: A novel moving object extraction algorithm based on a 3D self-organizing neural network to overcome the prominent challenges in microorganism video sequences, such as error bootstrapping, dynamic background, variable motion, physical deformation and noise obscured is proposed.
Abstract: Accurate detection and extraction of moving microorganisms from microscopic video streams is the first important step in biological wastewater treatment system. We propose a novel moving object extraction algorithm based on a 3D self-organizing neural network to overcome the prominent challenges in microorganism video sequences, such as error bootstrapping, dynamic background, variable motion, physical deformation and noise obscured. Firstly, we design a multilayer network topology instead of the traditional single-layer self-organizing map, which significantly improve the discrimination ability of moving objects. Secondly, new designed mechanisms related to background model initialization and adaptively update have effectively weakened the bootstrapping and ghost influences. Thirdly, we create buffer layers in neural network efficiently to resolve the dynamic background and variable motion problems. Finally, a simple Kalman predictor with constant coefficients has been constructed to tackle with the cases of microorganism being obscured or lost. Experimental results on real microscopic video sequences and comparisons with the state-of-the-art methods have demonstrated the accuracy of our proposed microorganism extraction algorithm.
2 citations
Cites background or methods from "An Efficient Neural Network Based B..."
...A multilayer perceptron (MLP) network is utilized to formulate the moving background model in [25], in which the input data are some hand-crafted features of pixels, and the output become background or targets markers....
[...]
...The typical ANNs include adaptive resonant theory (ART) network [24], multilayer perceptions (MLPs) [25] and self-organizing map (SOM) [26]....
[...]
References
More filters
04 Oct 2004
TL;DR: An algorithm to measure the vessel diameter to subpixel accuracy is presented, based on a two-dimensional difference of Gaussian model, which is optimized to fit aTwo-dimensional intensity vessel segment.
Abstract: Changes in retinal vessel diameter are an important sign of diseases such as hypertension, arteriosclerosis and diabetes mellitus. Obtaining precise measurements of vascular widths is a critical and demanding process in automated retinal image analysis as the typical vessel is only a few pixels wide. This paper presents an algorithm to measure the vessel diameter to subpixel accuracy. The diameter measurement is based on a two-dimensional difference of Gaussian model, which is optimized to fit a two-dimensional intensity vessel segment. The performance of the method is evaluated against Brinchmann-Hansen's half height, Gregson's rectangular profile and Zhou's Gaussian model. Results from 100 sample profiles show that the presented algorithm is over 30% more precise than the compared techniques and is accurate to a third of a pixel.
275 citations
TL;DR: A neural network architecture is proposed to form an unsupervised Bayesian classifier for this application domain that efficiently handles the segmentation in natural-scene sequences with complex background motion and changes in illumination.
Abstract: This paper presents a novel background modeling and subtraction approach for video object segmentation. A neural network (NN) architecture is proposed to form an unsupervised Bayesian classifier for this application domain. The constructed classifier efficiently handles the segmentation in natural-scene sequences with complex background motion and changes in illumination. The weights of the proposed NN serve as a model of the background and are temporally updated to reflect the observed statistics of background. The segmentation performance of the proposed NN is qualitatively and quantitatively examined and compared to two extant probabilistic object segmentation algorithms, based on a previously published test pool containing diverse surveillance-related sequences. The proposed algorithm is parallelized on a subpixel level and designed to enable efficient hardware implementation.
190 citations
25 Jun 2008
TL;DR: A competitive neural network is proposed to form a background model for traffic surveillance to enable efficient hardware implementation and to achieve real-time processing at great frame rates.
Abstract: This paper presents a neural background modeling based on subtraction approach for video object segmentation. A competitive neural network is proposed to form a background model for traffic surveillance. The unsupervised neural classifier handles the segmentation in natural traffic sequences with changes in illumination. The segmentation performance of the proposed neural network is qualitatively examined and compared to mixture of Gaussian models. The proposed algorithm is designed to enable efficient hardware implementation and to achieve real-time processing at great frame rates.
30 citations
TL;DR: It is shown that the cost function plays the most significant role in realizing high levels of performance, and the neural cost function's context-sensitive treatment of appearance, change of appearance and trajectory yield better tracking than a simple, explicitly designed cost function.
14 citations