Object detection in complex scenes (back)

Description

We propose an algorithm to detect moving objects in higly dynamic scenes with large changes in the background.

First, groups of pixels having similar motion and photometric features are extracted. For this first step only a sub-grid of image pixels is used to reduce computational cost to improve robustness to noise. The grid is regularly spread on the image. Each of its pixel is described by its position, motion and color. We introduce the use of p-value to validate optical flow estimates. An iterative algorithm for automatic bandwidth selection has been introduced (more) in order to apply a parameter free mean shift clustering algorithm.

In a second stage, segmentation of the object associated to a given cluster is performed in a MAP/MRF framework.

Our method is able to handle moving camera and several different motions in the background.

Results

Outdoor sequences

Water skier sequence ;

Object detection with mean shift clustering, Moving object segmentation and Motion detection

Drivers sequences

Original Image	Clustering	Mincut Segmentation

Reference

A. Bugeau, P. Pérez. Detection and segmentation of moving objects in highly dynamic scenes. Proc. Int. Conf. Computer Vision and Pattern Recog. (CVPR' 07), Minneapolis, MI, June 2007. (pdf)

A. Bugeau, P. Pérez. Detection and segmentation of moving objects in highly dynamic scenes.Technical report, INRIA, RR-6282, 2007.(pdf)