Real–Time Detection of Moving Obstacles from Mobile Platforms

In this paper we present a vision–based algorithm for the detection of moving obstacles in complex and unknown environments. The goal is to find moving objects from images captured by a mobile camera navigating together with a moving platform. One specific feature of our approach is that it does not need any information of the camera and hence works without camera calibration. Another advantage lies in the fact that it integrates motion separation and outlier detection into one statistical framework. Based on sparse point correspondences extracted from consecutive frame pairs, scene points are clustered into different classes by statistical analysis and modeling of the probability distribution function of the underlying motion characteristics. Experimental results based on several real–world video streams demonstrate the efficiency of our algorithm.