doi:10.1016/S0031-3203(01)00119-4 
Copyright © 2002 Pattern Recognition Society. Published by Elsevier Science B.V.
Estimating relative vehicle motions in traffic scenes
a Center for Automation Research, University of Maryland, College Park, MD 20742-3275, USA
b Computer Science Department, George Mason University, Fairfax, VA 22030-4444, USA
c Department of Computer Science, Technion, Israel Institute of Technology, Haifa 32000, Israel
Received 12 September 2000;
accepted 27 April 2001
Available online 28 February 2002.
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Abstract
Autonomous operation of a vehicle on a road calls for understanding of various events involving the motions of the vehicles in its vicinity. In this paper we show how a moving vehicle which is carrying a camera can estimate the relative motions of nearby vehicles. We show how to “smooth” the motion of the observing vehicle, i.e. to correct the image sequence so that transient motions (primarily rotations) resulting from bumps, etc. are removed and the sequence corresponds more closely to the sequence that would have been collected if the motion had been smooth. We also show how to detect the motions of nearby vehicles relative to the observing vehicle. We present results for several road image sequences which demonstrate the effectiveness of our approach.
Author Keywords: Traffic; Vehicle motion; Image stabilization; Darboux motion; Rate of approach
Fig. 1. The plane perspective projection image of P is F=f(X/Z,Y/Z,1); the weak perspective projection image of P is obtained through the plane perspective projection of the intermediate point P1=(X,Y,Zc) and is given by G=f(X/Zc,Y/Zc,1).
Fig. 2. (a–d) A selected frame from each of four sequences. Top: the input images. Middle: results of road detection. Bottom: results of vehicle detection.
Fig. 3. (a) and (b) Two images taken 1/15th of a second apart; (c) and (d) their normal flows. One can see the effects of bumps. In the first frame, the flow vectors point downward; in the second, they point upward.
Fig. 4. Derotation results for one frame from each of three sequences: (a) input frame; (b) normal flow; (c) rotational normal flow; (d) translational normal flow.
Fig. 5. Identification of distant image points in frames from two sequences.
Fig. 6. Frames 0, 30 and 60 of a sequence showing two vehicles accelerating. The normal flow results are shown below the corresponding image frames.
Fig. 7. Motion analysis results for the acceleration sequence. U,V,W are the scaled (by an unknown distance Zc−1) components of the relative translational velocity.
Fig. 8. Frames 5, 15, 25 and 35 of the van sequence. The normal flow results are shown below the corresponding image frames.
Fig. 9. Motion analysis results for the van sequence. U,V,W are the scaled (by an unknown distance Zc−1) components of the relative translational velocity.
Fig. 10. Frames 1, 14 and 26 of the Italian sequence. The normal flow results are shown below the corresponding image frames.
Fig. 11. Motion analysis results for the Italian sequence. U,V,W are the scaled (by an unknown distance Zc−1) components of the relative translational velocity.
Fig. 12. Frames 1, 25 and 48 of the Haifa sequence. The normal flow results are shown below the corresponding image frames.
Fig. 13. Motion analysis results for the Haifa sequence. U,V,W are the scaled (by an unknown distance Zc−1) components of the relative translational velocity.
Fig. 14. The Darboux frame moves along the path Γ which lies on the surface Σ.
Abstract
Purchase PDF (975 K)
E-mail Article
Add to my Quick Links
Cited By in Scopus (3)
Purchase PDF (5888 K)
