Development of person identification application for video surveillance
In the real world, CCTVs are implemented and allocated in public and private environment, to ensure public safety. However, it seems to like observing the video surveillance, to figure out a target, is not a simple task. It increases the human cost of video surveillance. Hence, person identification...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Year Project / Dissertation / Thesis |
| Published: |
2022
|
| Subjects: | |
| Online Access: | http://eprints.utar.edu.my/4665/1/fyp_CS_2022_SPC.pdf http://eprints.utar.edu.my/4665/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In the real world, CCTVs are implemented and allocated in public and private environment, to ensure public safety. However, it seems to like observing the video surveillance, to figure out a target, is not a simple task. It increases the human cost of video surveillance. Hence, person identification by the video surveillance application is developed in this project.
In this application, user inserts at least two CCTV videos and at most 4 CCTV videos, either in the public environment or private environment. Simultaneously, user inserts images of the target to be identified in these videos. In the end, the output is the videos with the green bounding boxes, which indicates as the target.
After inserting the necessary images and videos, it is time for the back-end process. There are three important processes in this application, such as person detection, person tracking, and person identification. First, person detection implemented YOLOv3. YOLOv3 is a common neural network algorithm for detection. It can detect 80 classes of objects such as a person, car, pot, and more. Therefore, it was set and adjusted to detect persons only. Second, person tracking is important to track the detected person. In this application, person tracking implemented the DeepSORT algorithm for tracking by the tracker. The trackers contained information such as track ID, class type, and bounding boxes. The information of the tracker is necessary for person identification. Third, person identification implemented the CNN model for training. After training, the videos with green bounding boxes are displayed, when the person is predicted as the target.
In conclusion, this person identification application for video surveillance improves the efficiency and accuracy of person identification in multiple videos, instead of physical surveillance by humans, which is resources inefficiency. |
|---|