LaBGen: A method based on motion detection for generating the background of a scene

Abstract

Given a video sequence acquired with a fixed camera, the generation of the stationary background of the scene is a challenging problem which aims at computing a reference image for a motionless background. For that purpose, we developed our method named LaBGen, which emerged as the best one during the Scene Background Modeling and Initialization (SBMI) workshop organized in 2015, and the IEEE Scene Background Modeling Contest (SBMC) organized in 2016. LaBGen combines a pixel-wise temporal median filter and a patch selection mechanism based on motion detection. To detect motion, a background subtraction algorithm decides, for each frame, which pixels belong to the background. In this paper, we describe the LaBGen method extensively, evaluate it on the SBI 2016 dataset and compare its performance with other background generation methods. We also study its computational complexity, the performance sensitivity with respect to its parameters, and the stability of the predicted background image over time with respect to the chosen background subtraction algorithm.We provide an open source C++ implementation at http://www.telecom.ulg.ac.be/labgen.

Is a Memoryless Motion Detection Truly Relevant for Background Generation with LaBGen?

“…In our previous works related to LaBGen [10], we have shown that, surprisingly, the frame difference background subtraction algorithm is the most effective motion detection, on average, for our background generation framework. Compared to other background subtraction algorithms, the main specificity of the frame difference algorithm is that it detects motion between two frames without relying on additional past frames.…”

“…The LaBGen background generation method, extensively described in [10], combines a pixel-wise median filter and a patch selection mechanism based on a motion detection performed by a background subtraction algorithm. It is composed of the five following steps, which are illustrated in Fig.…”

“…. , 201; and A with the frame difference, and 11 other background subtraction algorithms enumerated and described in [10]. The parameter values being considered for LaBGen-OF are the same for P, N and S; τ = 0.02, 0.03, .…”

Deep neural network concepts for background subtraction:A systematic review and comparative evaluation

“…Background generation [18,100,129] (also called background initialization [89,91] [ 92,176], background estimation [38,70], and background extraction [198]) regards the initialization of the background. Generally, the model is often initialized using the first frame or a background model over a set of training frames which contain or do [93] and LaBGen [105,106,107] that are based on robust PCA [20,21] and the robust estimation of the median, respectively. Practically, the main challenge is to obtain a first background model when more than half of the training contains foreground objects.…”

Markerless tracking of an entire insect colony

“…We additionally devised a method for detecting brood cells in a colony. We first used a background extraction method 71,72 applied to a range of images spanning 12 h. One background image was generated for each consecutive 12 h (Supplemental Movie M1). We then adopted our bee annotation tool (https://github.com/oist/DenseObjectAnnotation) to annotate center points of each brood cell in the generated background images.…”