Multi-Agent Approach for Occlusion Handling In Crowded Surveillance Video
| dc.contributor.author | Lee Beng Yong | |
| dc.date.accessioned | 2026-10-13T07:04:58Z | |
| dc.date.issued | 2018 | |
| dc.description | The task of object tracking is the key to the effective use of more advanced technologies such as event recognition in the surveillance video. To recognise an event, people and objects must be tracked. Tracking also enhances the performance of tasks such as crowd analysis or human identification. Among all existing object tracking methods, Continuously Adaptive Mean-shift (CAMShift) is one of the most popular methods adopted by many researchers. However, to use CAMShift to track an object, one should also verify that the chosen target representation is a sufficient discriminator for the application domain to ensure CAMShift can efficiently work in its application. This thesis investigates improvements to the performance of object tracking algorithms through improved target representation selection using a novel multi competitive agents system that consists of a manager agent, an evaluator agent and multiple CAMShift tracker agents. The interaction between the manager agent and the CAMShift tracker agents in the proposed is performed via the Call-For-Proposal protocol. Through the interaction, the CAMShift tracker agents in the system can compete among themselves to select an optimum target representation. The proposed algorithm is tested on lab simulation videos and shows significant improvement in tracking. Kalman filter is then added to the CAMShift tracker agents to solve full occlusion problem. When a tracked object is fully occluded, it is not visible in the video scenes. CAMShift tracker agent will not be able to retrieve any fully occluded object representation model from the scene, and therefore Kalman filter is used to predict the location of the tracked object based on spatial motion characteristic of the tracked object in previous frames. The proposed system is then tested on two famous benchmark surveillance video dataset, namely the PETS 2009 and CAVIAR. The results v show that the proposed system can track objects in crowded scenes even when a full occlusion occurs during the tracking process. | |
| dc.identifier.uri | https://scholarhub.unimas.my/handle/123456789/2984 | |
| dc.language.iso | English | |
| dc.publisher | UNIVERSITI MALAYSIA SARAWAK | |
| dc.relation.ispartofseries | Faculty of Computer Science and Information Technology | |
| dc.subject | Continuously Adaptive Mean-shift, Kalman Filter, Multi-Agents System Occlusion, Crowded Surveillance Video, Object Tracking | |
| dc.title | Multi-Agent Approach for Occlusion Handling In Crowded Surveillance Video | |
| dc.type | PhD |