detection

Moving Object Detection with Freely Moving Camera via
Background Motin Subtraction

Yuanyuan Wu^1,2, Xiaohai He¹, Truong Q. Nguyen²

¹College of Electronics and Information Engineering, Sichuan University, Chengdu, China
² ECE Department, University of California San Diego, La Jolla CA 92093

Abstract

Detection of moving objects in a video captured by a freely moving camera is a challenging problem in computer vision. Most existing methods often assume that the background can be approximated by dominant single plane/multiple planes, or impose significant geometric constraints of background, or utilize complex background/foreground probabilistic model. Instead, we propose a computationally efficient algorithm which is able to detect moving objects accurate and robust in a general 3D scene. This is obtained by casting the problem as a coarse-to-fine thresholding scheme on the particle trajectories in the video sequence. First, coarse foreground region is extracted by performing reduced singular value decomposition (RSVD) on multiple matrices which are built from bundles of particle trajectories. This process is able to handle motion parallax from 3D scene effectively. Next, the background motion of pixels in coarse foreground region are reconstructed by a fast inpainting method. After subtracting the background motion, the fine foreground is segmented out by proposed adaptive thresholding method which is capable of solving multiple moving objects scenarios. Finally, the detected foreground is further refined by the mean-shift segmentation method. Extensive qualitative and quantitative experiments against state-of-the-art methods verify the effectiveness of proposed method.

Publication

Wu, Yuanyuan, Xiaohai He, Truong Q. Nguyen. "Moving Object Detection with Freely Moving Camera via Background Motion Subtraction." (Under review)

Downloads

BMS Dataset

Some ground truth masks of moving objects in experimental video sequences.