A. Naftel (UK), J. Melo (UK, Portugal), A. Bernardino, and J. Santos-Victor (Portugal)
vehicle tracking, motion trajectory, lane detection, sceneinterpretation
This paper describes a clustering technique for detecting and classifying highway lanes using only vehicle motion trajectories. The detected lanes are described by low order polynomials and a directional indicator combined with simple distance metrics which permits classification of lanes into one of the following sets: primary, entry, exit or secondary. The approach taken is viewpoint independent and can be successfully applied to uncalibrated pan-tilt zoom cameras normally used in traffic surveillance. The algorithm does not require any a priori road feature extraction on static images. We show that for each viewpoint and classified lane, the vehicle density and lane changes can then be analysed. The clustering technique adopted here is shown to perform better than a standard histogram lane finding approach. Experimental results are presented that show the application of this approach to multiple views obtained by a PTZ camera monitoring the junction of two intersecting highways.
Important Links:
Go Back
