Moving Vehicle Tracking Using Multiple Uncalibrated Non-overlapped Cameras

Please enable javascript in your browser.

Shabani Nia, Elham | 2009

599 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 39891 (19)
University: Sharif University of Technology
Department: computer engineering department
Advisor(s): Kasaei, Shohreh
Abstract:
Multi-camera vehicle tracking is a necessary part of any video-based intelligent transportation systems (ITS) for extracting different traffic parameters such as link travel times and Origin/Destination (O/D) counts efficiently. In many applications it is needed to place traffic cameras disjoint from each other to cover wide areas. In this work we present a system for tracking vehicles in such network of cameras. The proposed system includes new ideas for handling inter-object occlusion as the most challenging part of single camera tracking phase. This approach is based on coding silhouette of moving objects before and after occlusion and separating occluded vehicles by finding longest common substring of chain codes. In addition, for improving accuracy of multi-camera tracking, a new feature based on neighbors’ relationships of a vehicle is formulated besides appearance and space-time features. In highways that vehicles tend to keep their relationships, using similarity of neighbors can be a helpful hint to find corresponding vehicles in different sites. A graph based approach has been used to track vehicles in the camera network. Experimental results show efficiency of the proposed methods
Keywords:
Disjoint View ; Multi Camera ; Vehicle Tracking ; Non Overlaped Cameras

No TOC