Abstract:Acknowledgment Moving object detection in dynamic scenes is a challenging problem due to the complex variations in the background such as the swaying leaves and rippers. Most of the existing methods usually cast the variation into background or foreground layer, and thus fail to precisely model the dynamic variation in an explicit manner. In this work, we formulate the moving object detection as tensor-based three layers decomposition problem: static background, moving object foreground, and dynamic background to address this issue. Specifically, we additionally introduce a data-driven discriminative prior embedded into the maximum a posterior (MAP) framework to explicitly capture the dynamic background, which can be satisfactorily represented by a two stream spatiotemporal convolutional neural network. As for the static background, we introduce the low-rank tensor prior to model the temporal correlation; For the moving object, we utilize its spatiotemporal continuity via the Markov random field. Compared with previous matrix-based methods, each term in our method is tensor-based which could better preserve the intrinsic spatiotemporal structure correlation in the video. In addition, our model could be further improved by involving a semantic prior as an adaptive attention. The proposed method is extensively evaluated on several benchmarks, and significantly outperforms state-of-the-art methods.

Learning complex background by multi-scale discriminative model

Learning Complex Background Based on Multi-scale Discriminative Model

Background modeling using mixture of Gaussians and Laplacian pyramid decomposition.

Moving Object Detection Based on Multi-layer Background Model

Detecting Moving Objects from Dynamic Background Combining Subspace Learning with Mixed Norm Approach

Multi-Channel Adaptive Mixture Background Model for Real-time Tracking.

Background Segmentation of Dynamic Scenes Based on Dual Model

Background Modeling by Exploring Multi-Scale Fusion of Texture and Intensity in Complex Scenes

A Multilayer-Based Framework for Online Background Subtraction with Freely Moving Cameras

Vision-Based Moving Objects Detection with Background Modeling

Moving object detection method based on complex fuzzy logic system

Foreground Detection: Combining Background Subspace Learning with Object Smoothing Model

A Multiple Layer Background Model for Foreground Detection

Efficient Background Model Based on Multi-Level Feedback for Video Surveillance

Complex Background Subtraction by Pursuing Dynamic Spatio-Temporal Models

A multiscale spatio-temporal background model for motion detection

Background Subtraction Through Multiple Life Span Modeling

Learning Dynamic Background for Moving Object Detection

A Double-Layer Model for Foreground Detection from Video Sequence

Background Segmentation Using Spatial-Temporal Multi-Resolution MRF.

A Framework for Foreground Detection in Complex Environments