Multiple person tracking and left luggage detection using appearance models | |
| Author | Gharti, Shashi |
| Call Number | AIT Thesis no.CS-10-05 |
| Subject(s) | Electronic surveillance Signal detection |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Computer Science, School of Engineering and Technology |
| Publisher | Asian Institute of Technology |
| Series Statement | Thesis ; no. CS-10-05 |
| Abstract | Increasing crime threatens our communities. To ensure security, surveillance cameras are placed in many public places. Video surveillance cameras require humans to watch the videos and monitor for suspicious events occurring in the scene. This is difficult for humans. They may miss criminal acts. Intelligent video surveillance systems have the potential to increase security and reduce the cost of human resources. This thesis focuses on finding suspicious bags or other objects left in public areas. The left luggage detection system is based on the assumption that if we can track people accurately then finding left luggage is easy. I explore a relatively simple general purpose system for tracking multiple people that also detects unattended objects left in the scene. The system consists of four modules: (a) foreground extraction (b) feature extraction (c) tracking and (d) search for unattended luggage. The system extracts foreground objects, removes noise from the foreground mask, then finds connected foreground components. It extracts features and tracks the objects throughout their lifetimes. I use color coherence vectors (ccvs) to deal with merges/splits and occlusions. The system continuously checks for any bag-like objects in the scene. It assumes that the motion of bag is less than that of humans and that bags are more compact than humans. If the system finds any suspicious object in the scene, it will continuously search for the object’s owner. If the owner is not found for sometime, the system raises and alarm. To evaluate the proposed approach, I performed experiments comparing my method to alternative approaches based on standard color histogram appearance models and appearance-free tracking using simple overlap heuristics for person tracking. My exper- iments show that color coherence vectors work better than the other two methods. It helps the system maintain the identity of people despite occlusions, merges and splits. It works specially well if a merge involves only two people. However, as the complexity of merges increases, the system may lose track of individual personal identities |
| Year | 2010 |
| Corresponding Series Added Entry | Asian Institute of Technology. Thesis ; no. CS-10-05 |
| Type | Thesis |
| School | School of Engineering and Technology (SET) |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Computer Science (CS) |
| Chairperson(s) | Dailey, Matthew N.; |
| Examination Committee(s) | Afzulpurkar, Nitin V.;Kiyoshi, Hond; |
| Scholarship Donor(s) | Asian Institute of Technology Fellowship; |
| Degree | Thesis (M.Eng.) - Asian Institute of Technology, 2010 |