Loading...

Multi-Object Tracking in Video using Graph Neural Networks

Hosseinzadeh, Mehran | 2023

106 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56145 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Rabiee, Hamid Reza
  7. Abstract:
  8. Multiple object tracking refers to the detection and following of target object classes in video sequences. In this task, all objects belonging to the target classes in the video are detected simultaneously in each frame, and a unique ID is assigned to each of them throughout the video. In recent years, the use of graph neural networks for solving this problem has received significant attention because these models are suitable tools for discovering and improving the relationships between objects in the scene, which can greatly assist in better object pairing. However, there are various challenges to using graph neural networks, the most important of which is the limitation of input graph size and heavy computational burden as the graph grows larger. This has led to less use of temporal relationships in videos in graph-based neural networks, resulting in reduced accuracy of these models for dense and highly occluded scenes. Therefore, the deficiency of existing graph-based methods is either the lack of a suitable mechanism to utilize temporal dependencies to overcome the difficulties and occlusions of crowded scenes or the heavy computational and training burden associated with implementing such a mechanism. Aiming to track all present objects simultaneously and accurately in each frame, especially in crowded scenes, this research has introduced a graph-based model that incorporates more temporal features into the construction of the graph used, and dynamically updates it over time using a graph neural network, with a relatively lightweight and without adding a significant computational burden. To achieve this goal, a component for constructing an adjacency matrix based on spatio-temporal features and another component for enforcing dynamicity are introduced. The results of applying the proposed model on two commonly used datasets in this field, as well as subsets containing denser images, demonstrate the effectiveness of the proposed approach and the improvement in the performance of graph-based models. Specifically, in denser scenes, the proposed model has shown an improvement of about 16% compared to baseline graph-based models, while also being computationally lightweight and efficient
  9. Keywords:
  10. Multiple Object Tracking ; Graph Neural Network ; Dynamic Neural Network ; Dynamic Target Tracking ; Object Detection

 Digital Object List

 Bookmark