Loading...

Unsupervised Abnormal Activity Detection and Description in Video

Hassani, Hesameddin | 2015

709 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 47385 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Sharif Khani, Mohammad; Gholampour, Iman
  7. Abstract:
  8. Nowadays, surveillance cameras are an integral part of the transportation management center. On one hand, developing of surveillance cameras around the cities and on the other hand, increasing costs of driving accidents, makes it necessary to use the automatic behavior analyzing systems on video sequences. The scientists’ focuses on detecting unusual events as one of the most important usage of video analysis, and recently, the algorithms of unusual action detecting had an impressive development by using the machine learning methods. In this research, the existing unusual event detection methods and in special, traffic unusual activities will place under scrutiny, and we suggest an algorithm for detecting such these activities that includes feature extraction and behavior representation and modeling, and also detection of abnormal events. We evaluate the suggested algorithm in different points of view, within many examinations on common datasets in this field and traffic video shots of Tehran streets, and show that suggested algorithm could detects every kind of unusual activities, successfully. Results show the superiority of the proposed solution to other methods are known today, as far as accuracy increased by 10.48% and 8.81% about unusual event detection in two well-known datasets. In term of behavior modeling, suggested model compared with other existing models, that could confirms its benefits versus others and in this part we have been reached to 9.05% and 2.67% improvement in two mentioned datasets
  9. Keywords:
  10. Surveillance System ; Machine Learning ; Feature Extraction ; Constitutive Modeling ; Abnormal Event Detection ; Surveillance Cameras

 Digital Object List

 Bookmark

No TOC