Abstract-Background subtraction is usually based on lowlevel or hand-crafted features such as raw color components, gradients, or local binary patterns. As an improvement, we present a background subtraction algorithm based on spatial features learned with convolutional neural networks (ConvNets). Our algorithm uses a background model reduced to a single background image and a scene-specific training dataset to feed ConvNets that prove able to learn how to subtract the background from an input image patch. Experiments led on 2014 ChangeDetection.net dataset show that our ConvNet based algorithm at least reproduces the performance of state-of-the-art methods, and that it even outperforms them significantly when scene-specific knowledge is considered.

We introduce the notion of semantic background subtraction, a novel framework for motion detection in video sequences. The key innovation consists to leverage object-level semantics to address the variety of challenging scenarios for background subtraction. Our framework combines the information of a semantic segmentation algorithm, expressed by a probability for each pixel, with the output of any background subtraction algorithm to reduce false positive detections produced by illumination changes, dynamic backgrounds, strong shadows, and ghosts. In addition, it maintains a fully semantic background model to improve the detection of camouflaged foreground objects. Experiments led on the CDNet dataset show that we managed to improve, significantly, almost all background subtraction algorithms of the CDNet leaderboard, and reduce the mean overall error rate of all the 34 algorithms (resp. of the best 5 algorithms) by roughly 50% (resp. 20%). Note that a C++ implementation of the framework is available at http://www.telecom.ulg.ac.be/semantic.

The estimation of the background image from a video sequence is necessary in some applications. Computing the median for each pixel over time is effective, but it fails when the background is visible for less than half of the time. In this paper, we propose a new method leveraging the segmentation performed by a background subtraction algorithm, which reduces the set of color candidates, for each pixel, before the median is applied. Our method is simple and fully generic as any background subtraction algorithm can be used. While recent background subtraction algorithms are excellent in detecting moving objects, our experiments show that the frame difference algorithm is a technique that compare advantageously to more advanced ones. Finally, we present the background images obtained on the SBI dataset, which appear to be almost perfect.

No abstract

This paper proposes a new pixel-based background subtraction technique, applicable to range images, to detect motion. Our method exploits the physical meaning of depth information, which leads to an improved background/foreground segmentation and the instantaneous suppression of ghosts that would appear on color images.In particular, our technique considers certain characteristics of depth measurements, such as failures for certain pixels or the non-uniformity of the spatial distribution of noise in range images, to build an improved pixel-based background model. Experiments show that incorporating specificities related to depth measurements allows us to propose a method whose performance is increased with respect to other stateof-the-art methods.