COROLA: A sequential solution to moving object detection using low-rank approximation

Abstract

Extracting moving objects from a video sequence and estimating the background of each individual image are fundamental issues in many practical applications such as visual surveillance, intelligent vehicle navigation, and traffic monitoring.Recently, some methods have been proposed to detect moving objects in a video via low-rank approximation and sparse outliers where the background is modeled with the computed low-rank component of the video and the foreground objects are detected as the sparse outliers in the low-rank approximation. Many of these existing methods work in a batch manner, preventing them from being applied in real time and long duration tasks. To address this issue, some online methods have been proposed; however, existing online methods fail to provide satisfactory results under challenging conditions such as dynamic background scene and noisy environments. In this paper, we present an online sequential framework, namely contiguous outliers representation via online low-rank approximation (COROLA), to detect moving objects and learn the background model at the same time. We also show that our model can detect moving objects with a moving camera. Our experimental evaluation uses simulated data and real public datasets to demonstrate the superior performance of COROLA to the existing batch and online methods in terms of both accuracy and efficiency.

Moving Object Detection Under Discontinuous Change in Illumination Using Tensor Low-Rank and Invariant Sparse Decomposition

“…The third group of methods also imposed the connectivity constraint on S [32,30,35,21,31,17] using other formulations than the second group. For example, Liu et al [17] attempted to use a structured sparsity norm [19] and a motion saliency map, to improve the accuracy of moving object segmentation under sudden illumination changes.…”

A Fusion Framework for Camouflaged Moving Foreground Detection in the Wavelet Domain

“…In [36], a framework named detecting contiguous outliers in the low-rank representation (DECOLOR) was proposed where the object detection and background learning are integrated into a single optimization process solved by an alternating algorithm. In [37], an online sequential framework named contiguous outliers representation via online low-rank approximation (COROLA) was proposed to detect moving objects and learn the background model at the same time. It works iteratively on each image of the video to extract foreground objects by exploiting the sequential nature of a continuous video of a scene where the background model does not change discontinuously and can therefore be obtained by updating the background model learned from preceding images.…”

“…Several popular and representative methods are compared, including "MOG2" [20], "FuzzyChoquetIntegral" [80], "LBAdaptiveSOM" [81], "Mul-tiLayerBGS" [83], "SuBSENSE" [82], "PBAS" [31], "DE-COLOR" [36], "COROLA" [37]. The "LSD" method [35] is similar to "DECOLOR" [36] and "COROLA" [37] methods in terms of low-rank representation and performance on the camouflaged videos and so is not further included in the comparison. The results of these methods are obtained based on the bgslibrary [84], [85] and the LRSLibrary [86], [87].…”

“…The time is for processing a frame of resolution 640 * 384 and is obtained on a personal computer (i7-4790 CPU with 32GB memory). Note that "DECOLOR" [36] and "COROLA" [37] are implemented in Matlab while others are implemented in C++. FWFC takes more time than MOG2 [20] but less time than most of the existing methods as listed in Table III.…”

Fast Shadow Detection from a Single Image Using a Patched Convolutional Neural Network

“…Dealing with shadows is one of the most fundamental issues in image processing, computer vision and robotics. Shadows are omnipresent in outdoor applications and must be taken into account in the solutions to standard computer vision and robotics problems such as image segmentation [1], change detection [2], place recognition [3], background subtraction [4], [5], visual robot localization [6], [7] and navigation [8]. Unfortunately in most of these cases images are strongly influenced by shadow at different times of a day, making it difficult to interpret or understand a scene.…”