Le, Yi
(2024)
Dynamic Neural Fields Processing Using Temporal Dynamic Vision Sensing and Adaptive Threshold for Motion Tracking.
Master's Thesis, University of Pittsburgh.
(Unpublished)
Abstract
Dynamic Neural Fields (DNF) are models that mimic brain functions, which are good at processing information over time and making decisions in changing environments. They can create stable activity patterns for tasks like tracking moving objects and making quick decisions. Similarly, Dynamic Vision Sensors (DVS) have been used in past research for their ability to quickly capture moving objects in scenes. This paper is based on DNF that utilizes DVS as its input to track selected objects. The algorithm enhances the grayscale values of the chosen object, effectively setting the grayscale values of other parts of the image to zero, which is represented as the black part. However, interfering objects close to the selected object might also receive a high value and thus appear in the output. The original approach employed a sigmoidal function to introduce non-linearity to the image, aiming to differentiate between the target and interfering objects. This method, though, has its limitations: for larger objects, the non-linear features might result in the elimination of parts of the object itself, compromising the accuracy of the tracking process.
This work proposes the use of an adaptive threshold applied to the linear function, instead of the sigmoidal function, to distinguish objects. By directly implementing a simple algorithm, one can find out the value of the elements that are related to the threshold and the threshold can be calculated with the relevant elements. In the experiments conducted, by comparing the deviations in the object centers, the proposed method achieved an error reduction of approximately 92% compared to the use of a sigmoidal function. This reduction reveals that adaptive threshold can not only eliminate the existing interference objects but also avoid further bias in tracking, making the tracking performance better than using the sigmoidal function. Moreover, the proposed method keeps the performance stable when the tracking object becomes large, implying a wider range of applicability.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
3 June 2024 |
| Date Type: |
Publication |
| Defense Date: |
29 March 2024 |
| Approval Date: |
3 June 2024 |
| Submission Date: |
7 April 2024 |
| Access Restriction: |
No restriction; Release the ETD for access worldwide immediately. |
| Number of Pages: |
49 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
Swanson School of Engineering > Electrical and Computer Engineering |
| Degree: |
MS - Master of Science |
| Thesis Type: |
Master's Thesis |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
Motion Tracking, Dynamic Neural Field, Dynamic Vision Sensor |
| Date Deposited: |
03 Jun 2024 14:53 |
| Last Modified: |
03 Jun 2024 14:53 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/46047 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}