Tracking Algorithm Performance Evaluation on Embedded Processors
Visual Object tracking problem has been studied by the computer vision community for several decades. However, it remains a challenging task to design an efficient and robust visual object tracking system for all the practical real- world applications. Further, there are several factors which a
2025-06-28 16:36:26 - Adil Khan
Tracking Algorithm Performance Evaluation on Embedded Processors
Project Area of Specialization Artificial IntelligenceProject SummaryVisual Object tracking problem has been studied by the computer vision community for several decades. However, it remains a challenging task to design an efficient and robust visual object tracking system for all the practical real- world applications. Further, there are several factors which affect the performance of the object tracker such as illumination variations, scale variations, occlusions, deformations, motion blur, rotations, and low resolutions.
Object tracking serves as a fundamental part in many real-world applications, such as security surveillance, air traffic control, automated manufacturing and sports analysis. For time critical tasks such as traffic safety, real-time constraint is usually applied. So there is very need of an optimized tracking algorithm that is accelerated by hardware and that has the minimum time delay between the input stream of frames and the output stream.
Our goal is the implementation and validation of the algorithms for precision object tracking. The implemented algorithms will be independent of the size of the target and less sensitive to the intensity changes of the target.
Different object tracking algorithm will be studied and their performance will me evaluated by implementing on different embedded processor platform to find an optimized and efficient algorithm that can meet the industry standard and can be used in any relevant real world application.
Project ObjectivesThe main aims and objective of Tracking Algorithm Performance Evaluation on Embedded Processor are given below:
- Embedded Processor Platform Hardware/Software understanding.
- Video Capture/Display through PAL/camera link interface.
- Algorithm optimization in MATLAB.
- Hardware implementation of the selected algorithms on different Hardware platform
- Performance evaluation/comparison of the selected algorithms.
In the object tracking task, the tracker needs to estimate the location of the bounding box of the object within each frame of the video sequence. These methods model the target localization problem as a classification problem.
Initially, the target’s bounding box or tracking window is selected either manually or automatically. In the manual case, a supervisor marks the rectangle boundary (left, top, right, and bottom) of the target object of interest. Whereas, in the automatic setup as in the proposed system, the prior object detection method outputs the boundary of the target. We use discriminative correlation filter for tracking the object. The object tracking task is performed using these three steps:
- after selecting object boundary, a correlation filter is trained using image patch cropped from the first frame;
- in the subsequent frames, the target is tracked by correlating the trained correlation filter over a search window. The window location which gives maximum correlation output is marked as the new location of the target;
- finally, based on this new location, an online update of the correlation filter is performed. Steps 2 and three are repeated for all the frames in a video sequence.
This project can be helpful in following applications:
-
Hardware acceleration for edge computation.
-
Used by Educational Institutions for Research purposes.
-
Security Organization for tracing suspicious activities
-
Safe Cities for preventing theft, illegal acts.
-
Self Driving Cars
The Final Deliverables of the project will be as follows:
-
An optimized algorithm for visual object tracking implemented on two different embedded processors, that will be able to take an input of real time scenario from camera being processed through embedded processor and displaying given object being tracked by the time.
-
A detailed report about the performance evaluation of different Object Tracking Alogrithms on each of the given embedded processor platform
| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 80000 | |||
| Nvidia Jetson Nano | Equipment | 1 | 22000 | 22000 |
| Jetson Nano Camera | Equipment | 1 | 7000 | 7000 |
| Raspberry Pi 4 Model B 4GB | Equipment | 1 | 12500 | 12500 |
| Raspberry Pi Camera | Equipment | 1 | 4500 | 4500 |
| Intel Neurol Compute Stick 2 | Equipment | 1 | 24000 | 24000 |
| Misc | Miscellaneous | 1 | 10000 | 10000 |