Multiple Target Tracking for intelligent headlight control

This web page contains the results in video form of the paper

Multiple Target Tracking of intelligent headlight control
Jose C. Rubio, Joan Serrat, Antonio M. López and Daniel Ponsa
submitted to the 13th IEEE Int. Conf. on Intelligent Transportation Systems, 2010

Abstract

Intelligent vehicle lighting systems aim at automatically regulate the headlights' beam angle so as to illuminate as much of the road ahead as possible, while avoiding dazzling other drivers. A key component of such a system is a classifier able to distinguish blobs due to vehicles' head and rear-lights from those originating from road lamps and reflective elements like poles and traffic signs. The most challenging cases are faint and tiny blobs corresponding to quite distant vehicles which disappear and reappear now and then.

We address the problem by tracking blobs in order to 1) obtain more feature measurements per blob along its track, 2) compute motion features, which we deem relevant for the classification and 3) enforce its temporal consistency. This paper focuses on the problem of constructing blob tracks, which is actually one of multiple target tracking, but under special conditions: we have to deal with frequent occlusions as well as blob splitings and mergings.

Sequences

Here we show the target tracking for sequences containing several target occlusions, mergings and splittings. Videos were recorded by a camera with a CMOS image sensor from Aptima Imaging of 752x480 pixels of resolution. The lens has a 40 degrees angular field of view.

The following videos show the tracking results in the following way:

• Track colors denote their identity, so different colors mean different targets.
• However, a merging of two different targets produces a new color in the resulting track.
• At each frame, detections associated to a track in the past ten frames are shown, in case they exist.
• The right image presents a vertical projection of the tracks, x position against time. There, a change in the color of the track does not necessarely mean a track mislabeling: the vertical projection of the tracks may hide blobs which are vertically aligned. Next figure shows the projection procedure.
  
  
  
  

Sequence A

Several splittings and mergings of pole reflections near the camera.
Distant tracks, produced mainly by reflections on road structural elements.
Long occlusions (4 frames) correctly recovered.
Results: 92% correct labels, 57/96 occlusions recovered, 8/21 Mergings detected, and 17/25 splittings detected.

Sequence B

Several splittings and mergings of pole reflections near the camera.
Distant tracks, produced mainly by reflections on road structural elements.
Multiple occlusions of distant and small targets.
Results: 89% correct labels, 33/57 occlusions recovered, 5/8 mergings detected, and 6/11 splittings detected.

Sequence C

Very long tracks correponding to two distant vehicle tail-lights fully tracked.
Few occlusions, splittings and mergings.
Results: 94% correct labels. 5/8 occlusions recovered, 2/5 mergings detecte, and 4/6 splittings detected.

Sequence D

Several splittings and mergings of pole reflections near the camera.
Distant tracks, produced mainly by reflections on road structural elements.
Results: 91% correct labels, 7/9 occlusions recovered, 3/7 mergings derected, and 6/10 splittings detected.

Sequence E

Several splittings, mergings and occlusions produced by the oversegmentation and reflections of a vehicle approaching to the camera.
Results: 88% correct labels, 20/28 occlusions recovered, 5/11 mergings detected, and 7/12 splittings detected.

Last update: Apr 29, 2010