Implementing and Evaluating a Tracking-By-Detection Algorithm for a Camera Monitoring System
- In this paper, we designed, implemented, and tested a special surveillance camera system based on a combination of classical image processing algorithms. The system’s sub-objective consists of tracking experimental vehicles driving on a defined trajectories (Rail) in real time. Furthermore, it analyzes the scene to collect additional vehicles & rail-related information. The system then uses the gathered data to reach its main objective which confines oneself in independently predicting vehicles collision. Consequently, we propose a hybrid method of detecting and tracking ATLAS-vehicles efficiently. To detect the vehicle at the beginning of the video, periodically every n-frame, and in the case where the tracked vehicle has been lost, we used Histogram Back-Projection. By contrast, Kernelized correlation filter is used to track the detected vehicles. Combining these two methods provides one of the best trade-offs between accuracy and speed even on a single processing core. The proposed method achieves the best performance compared with three different approaches on a custom dataset.
Author: | Khaled Jbaili, Jan Thomanek |
---|---|
DOI: | https://doi.org/10.48446/opus-12332 |
ISSN: | 1437-7624 |
Parent Title (German): | 26. Interdisziplinäre Wissenschaftliche Konferenz Mittweida |
Publisher: | Hochschule Mittweida |
Place of publication: | Mittweida |
Document Type: | Conference Proceeding |
Language: | English |
Year of Completion: | 2021 |
Publishing Institution: | Hochschule Mittweida |
Release Date: | 2021/05/21 |
Tag: | Object Detection and Tracking; Planar Homography |
Issue: | 002 |
Page Number: | 10 |
First Page: | 193 |
Last Page: | 204 |
Open Access: | Frei zugänglich |
Licence (German): | Urheberrechtlich geschützt |