CORT: Class-Oriented Real-time Tracking for Embedded Systems

Edoardo Cittadini,Alessandro De Siena,Giorgio Buttazzo

2024-07-20

Abstract:The ever-increasing use of artificial intelligence in autonomous systems has significantly contributed to advance the research on multi-object tracking, adopted in several real-time applications (e.g., autonomous driving, surveillance drones, robotics) to localize and follow the trajectory of multiple objects moving in front of a camera. Current tracking algorithms can be divided into two main categories: some approaches introduce complex heuristics and re-identification models to improve the tracking accuracy and reduce the number of identification switches, without particular attention to the timing performance, whereas other approaches are aimed at reducing response times by removing the re-identification phase, thus penalizing the tracking accuracy. This work proposes a new approach to multi-class object tracking that allows achieving smaller and more predictable execution times, without penalizing the tracking performance. The idea is to reduce the problem of matching predictions with detections into smaller sub-problems by splitting the Hungarian matrix by class and invoking the second re-identification stage only when strictly necessary for a smaller number of elements. The proposed solution was evaluated in complex urban scenarios with several objects of different types (as cars, trucks, bikes, and pedestrians), showing the effectiveness of the multi-class approach with respect to state of the art trackers.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper aims to address the issues of real-time performance and resource efficiency in the field of Multiple Object Tracking (MOT), especially in embedded systems. Specifically, the paper proposes a new method called CORT (Class-Oriented Real-time Tracking) for handling multi-class object tracking in complex scenarios.

CORT mainly addresses the following issues:
1. **Execution Time Optimization**: By decomposing the Hungarian Algorithm into multiple class-based subproblems, the overall matching phase's time consumption is reduced. This decomposition allows for parallel processing of different classes of objects, thereby improving resource utilization efficiency.
2. **Resource Efficiency Improvement**: This method not only reduces the number of similarity scores that need to be computed but also lowers memory requirements, enhancing both time and space scalability.
3. **Maintaining Tracking Performance**: Achieves shorter and more predictable execution times without sacrificing tracking accuracy. This is accomplished through a two-stage cascade matching process, where the first stage uses simple Intersection over Union (IoU) for quick matching, and the second stage employs an appearance-based model for matching the remaining difficult-to-match objects.

The key contributions of the paper include:
- Proposing a class-based parallel matching problem formulation method by segmenting the Hungarian matrix by class, allowing each subproblem to be processed in parallel.
- Designing a new cascade matching process that combines IoU and appearance models to reduce computational load and execution time.
- Evaluating the method in complex urban scenarios, demonstrating the effectiveness of the multi-class approach and comparing it with existing trackers.

Experimental results show that in various types of scenarios, especially when objects are evenly distributed across different classes, CORT can significantly reduce execution time and memory consumption. Additionally, compared to existing technologies like BYTEtrack, CORT achieves faster processing speeds while maintaining or enhancing tracking performance.