Comparison of different optical flow methods on optical flow based out-of-distribution detection
In Cyber Physical Systems (CPS), object detection is a crucial aspect that ensures the safe and robust autonomous operation of such systems. However, studies have suggested that these object detection algorithms are susceptible to anomalies that are falsely classified. Consequently, there is a need...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2022
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/162920 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | In Cyber Physical Systems (CPS), object detection is a crucial aspect that ensures the safe and robust autonomous operation of such systems. However, studies have suggested that these object detection algorithms are susceptible to anomalies that are falsely classified. Consequently, there is a need to develop high-performing Out-of- Distrubution (OOD) detection algorithms for such systems.
The use of optical flow as a motion detection algorithm has historically been well- documented, and works have been done to incorporate optical flow into Variational Autoencoders (VAEs). This project proposes another method of OOD detection by incorporating the Lucas-Kanade model into a VAE. This proposed model is compared against an existing optical flow-basedd VAE that was implemented with the Farneback algorithm, and the performance and accuracy of both models were tested with a dataset that was self-collected. |
|---|