Vehicle identification from road side camera
The development of machine learning and the prosperity of the convolution neural network has brought new possibilities to the field of computer vision. Computer vision applications like object detection, tracking, etc. can be deployed in people’s daily life to enhance work efficiency and security. F...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2020
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/139642 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | The development of machine learning and the prosperity of the convolution neural network has brought new possibilities to the field of computer vision. Computer vision applications like object detection, tracking, etc. can be deployed in people’s daily life to enhance work efficiency and security. For this final year project, we focused on one common machine learning task called multi-object tracking, particularly from surveillance cameras. This technique can be utilized to monitor traffic flows and be useful in future smart city applications. In the project, we deployed a state-of-the-art person tracking algorithm called Simple Online and Realtime Tracking with a Deep Association Metric (Deep SORT) on the UA-DETRAC vehicle tracking dataset. Furthermore, we made some modifications to the original algorithm in both tracking strategy and re-identification techniques to achieve better tracking accuracy. As a result, we managed to improve the tracking accuracy from 16.8% to 19.4% and 25.4% to 30.4% for Evolving Box and Mask R-CNN detection inputs respectively. Our tracker ended up to be one of the state-of-the-art trackers on the dataset. |
|---|