Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 47991 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Gholampoor, Iman; Sharif Khani, Mohammad
- Abstract:
- In this work we present a surveillance video retrieval system based on Topic Models. We’ve shown that employing Dynamic Programming improve the effectiveness of Topic Model based retrival. In the other hand proposed method has the accuracy near to the low-level features based methods. Lightweight database is the major advantageous of propsed method over the low-level features based methods. In our work storage space occupied by database decreases from 42 MB to only 0.4 MB for the mit dataset. Moreover lightweight database strikingly speeded up the retrieval process, for example retrieval process in proposed method is about 24 times faster than low-level features based systems. Furthermore in this work two methods has proposed to make it possible to query low-occurrence activities. In the first method new topics are created by using leaned topics without any learning and unsupervised. In the second method we consider a secondary model in order to cover activities uncaptured by the main model. Experiments demonstrate effectiveness of proposed methods in retrieving low-occurrence activities. Finally for better interaction with the retrieval system an application is developed which supports various queries defined in the proposed query language. Also feature extraction is accelerated and approach to real time using GPU
- Keywords:
- Dynamic Programming ; Content Based Retrieval ; Topic Model ; Query ; Lightweight Database ; Fast Retrieval ; Low Occurence Activities
Digital Object List-
محتواي کتاب - view
Bookmark