Pedestrian detection and tracking | |
| Author | Bajracharya, Prina |
| Call Number | AIT Thesis no.IM-08-06 |
| Subject(s) | Electronic surveillance Pedestrians |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Information Management, School of Engineering and Technology |
| Publisher | Asian Institute of Technology |
| Series Statement | Thesis ; no. IM-08-06 |
| Abstract | Crime and terrorism have become major threats to human lives. These days we hear and read about the acts of terrorism almost every day. Due to increase in crime, there is an acute need to develop robust security systems. Video surveillance systems can be one of the effective solutions to prevent growing global terrorism and other crimes. Nowadays video surveillance systems are used in order to ensure the security of hu¬man lives. Both open source and commercial surveillance systems are readily available and cameras are also reasonably priced. Surveillance system can be either automatic or manual. One of the drawbacks of manual video surveillance systems is that they require additional human resources. For instance, it requires a human operator to continuously monitor a video to alert the security officer in the case of a suspicious event. This drawback of manual video surveillance can be overcome by implementing an automated surveillance system. In order to build a surveillance system that works in every situation, we require various modules. Some examples are (a) a observation and recording module, (b) a motion detection module, (c) a pedestrian detection and tracking module, (d) an identification module, and (e) a behavior profiling module. In this research I focus on one of the above mentioned modules, a pedestrian detection and tracking module. My pedestrian detection and tracking module consists of three sub modules: (a) a training module, (b) a detection module, and (c) a tracking module. In the training process, meaningful features are extracted which are then combined together to form a classifier that can discriminate pedestrians and non-pedestrians in a given scene. In the detection module, the trained classifier is used to classify image patches. I perform experiments on pedestrian tracking using three methods: (a) estimating the position, size, and velocity of pedestrians using Kalman filters, (b) linking detection by selecting the best-matching color coherence vectors (CCV), and (c) check the overlap between two sequential aetections. From the experiment, the results show that in simple scenarios containing a single pedestrian, both the detection and tracking algorithm work effectively. However, as the situation gets complex as in multiple pedestrian scenarios containing occlusion and crossover, the detection rate drops drastically, affecting the tracking results. |
| Year | 2008 |
| Corresponding Series Added Entry | Asian Institute of Technology. Thesis ; no. IM-08-06 |
| Type | Thesis |
| School | School of Engineering and Technology (SET) |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Information Management (IM) |
| Chairperson(s) | Dailey, Matthew N.; |
| Examination Committee(s) | Haddawy, Peter;Manukid Parnichkun; |
| Scholarship Donor(s) | Asian Institute of Technology Fellowship; |
| Degree | Thesis (M.Eng.) - Asian Institute of Technology, 2008 |