Multiple object tracking using deep learning
Multiple Object Tracking (MOT) is widely used in various fields, such as traffic flow monitoring and crowd density estimation. Related technologies can increase productivity by enabling fully automated object recognition and mitigating the risks associated with human error. It is particularly import...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2024
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/177601 |
| _version_ | 1824455408160342016 |
|---|---|
| author | Zhang, Bohan |
| author2 | Yap Kim Hui |
| author_facet | Yap Kim Hui Zhang, Bohan |
| author_sort | Zhang, Bohan |
| collection | NTU |
| description | Multiple Object Tracking (MOT) is widely used in various fields, such as traffic flow monitoring and crowd density estimation. Related technologies can increase productivity by enabling fully automated object recognition and mitigating the risks associated with human error. It is particularly important in public transportation. Deep learning makes it a powerful tool for handling complex tasks. Therefore, in this project, we aim to develop deep learning-based multi-object tracking method for the application of public transportation.
This project compares two algorithms, Deep Simple Online and Realtime Tracking (DeepSORT) and Strong Simple Online and Realtime Tracking (StrongSORT). DeepSORT represents an early deep learning-based object tracking model, while StrongSORT builds upon this foundation, introducing enhancements like AFLink and Gaussian Smooth Interpolation (GSI) to improve tracking accuracy further. The differences are assessed through their performance on the MOT16, MOT17 and MOT20 datasets from the MOT Challenge and the road monitoring video recorded from above by surveillance camera. The experimental results show that StrongSORT consistently outperforms DeepSORT on all test datasets. Even under conditions of high congestion, StrongSORT can still achieve reasonable accuracy for multi-object tracking tasks. The performance gap is more significant in dense tracking scenarios. |
| first_indexed | 2025-02-19T03:37:44Z |
| format | Thesis-Master by Coursework |
| id | ntu-10356/177601 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2025-02-19T03:37:44Z |
| publishDate | 2024 |
| publisher | Nanyang Technological University |
| record_format | dspace |
| spelling | ntu-10356/1776012024-05-31T15:49:57Z Multiple object tracking using deep learning Zhang, Bohan Yap Kim Hui School of Electrical and Electronic Engineering EKHYap@ntu.edu.sg Computer and Information Science Object detection Multiple object tracking DeepSORT StrongSORT Multiple Object Tracking (MOT) is widely used in various fields, such as traffic flow monitoring and crowd density estimation. Related technologies can increase productivity by enabling fully automated object recognition and mitigating the risks associated with human error. It is particularly important in public transportation. Deep learning makes it a powerful tool for handling complex tasks. Therefore, in this project, we aim to develop deep learning-based multi-object tracking method for the application of public transportation. This project compares two algorithms, Deep Simple Online and Realtime Tracking (DeepSORT) and Strong Simple Online and Realtime Tracking (StrongSORT). DeepSORT represents an early deep learning-based object tracking model, while StrongSORT builds upon this foundation, introducing enhancements like AFLink and Gaussian Smooth Interpolation (GSI) to improve tracking accuracy further. The differences are assessed through their performance on the MOT16, MOT17 and MOT20 datasets from the MOT Challenge and the road monitoring video recorded from above by surveillance camera. The experimental results show that StrongSORT consistently outperforms DeepSORT on all test datasets. Even under conditions of high congestion, StrongSORT can still achieve reasonable accuracy for multi-object tracking tasks. The performance gap is more significant in dense tracking scenarios. Master's degree 2024-05-29T04:56:38Z 2024-05-29T04:56:38Z 2024 Thesis-Master by Coursework Zhang, B. (2024). Multiple object tracking using deep learning. Master's thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/177601 https://hdl.handle.net/10356/177601 en application/pdf Nanyang Technological University |
| spellingShingle | Computer and Information Science Object detection Multiple object tracking DeepSORT StrongSORT Zhang, Bohan Multiple object tracking using deep learning |
| title | Multiple object tracking using deep learning |
| title_full | Multiple object tracking using deep learning |
| title_fullStr | Multiple object tracking using deep learning |
| title_full_unstemmed | Multiple object tracking using deep learning |
| title_short | Multiple object tracking using deep learning |
| title_sort | multiple object tracking using deep learning |
| topic | Computer and Information Science Object detection Multiple object tracking DeepSORT StrongSORT |
| url | https://hdl.handle.net/10356/177601 |
| work_keys_str_mv | AT zhangbohan multipleobjecttrackingusingdeeplearning |