Visual collision avoidance

Matthews, Neil David (1998) Visual collision avoidance. University of Southampton, Doctoral Thesis.

Abstract

Collision avoidance systems offer the possibility of significantly reducing the costs of road and specifically motorway driving, in terms of both accidents and their associated effects. Additionally, similar schemes such as autonomous intelligent cruise control (AICC) also offer the possibility of increased road capacity.

Accidents at night are relatively more frequent and severe, and thus this work initially concentrates on the development of techniques and algorithms to automatically detect potential night-time collisions. Hence, this thesis considers the design and development of a complete processing chain, from initial image acquisition, through the recovery of trajectory measures describing the threat or otherwise posed by other road vehicles.

Since lights are the dominant features in night imagery, a feature extraction algorithm has been developed to locate and extract vehicle tail-lights from input imagery. Trajectories are extracted by tracking such vehicle cues temporally, which may then be passed to a set of trajectory analysis processes.

The trajectory analysis initially considers area-based recession rates recovered from single regions. However, due to the low signal to noise ratio in the recovered areas and recession areas, a novel pairing technique is developed enabling a subsequent pair-wise trajectory analysis to be performed, which allows significantly improved recession rates to be recovered. A additional clustering technique is also developed which allows pairings consistent with individual vehicles to be aggregated, substantially reducing the number of recovered targets.

These techniques developed for night driving are further developed and extended to daylight scenarios, by the application of a novel combined vehicle detection and classification system.

This record has no associated files available for download.

Contributors

Download statistics