Baratov, Farrukh (2022) Towards safer autonomous vehicles: Applying uncertainty quantification to pedestrian detectors. Bachelor's Thesis, Artificial Intelligence.
|
Text
bAI_2022_BaratovF.pdf Download (2MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (123kB) |
Abstract
Object detection networks have progressed at a rapid pace in recent years. However, detectors still suffer from many issues, such as a lack of network transparency and overconfidence in their predictions. Putting too much trust into unreliable predictions can have fatal consequences in safety-critical applications such as pedestrian detection for autonomous vehicles. A measure of network uncertainty can be implemented in order to give insight into the reliability of the detector. This study investigates how Monte Carlo Dropout, a method of uncertainty quantification, can be utilized to improve the performance of a Single-Shot Detector (SSD) network trained to detect pedestrians. We find that, based on the test dataset mAP scores of the detectors, the MC-Dropout model is able to improve its precision by about 7.47% compared to the baseline when it predictions with high uncertainty are suppressed.
| Item Type: | Thesis (Bachelor's Thesis) |
|---|---|
| Supervisor name: | Valdenegro Toro, M.A. |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Bachelor's Thesis |
| Language: | English |
| Date Deposited: | 16 Aug 2022 09:09 |
| Last Modified: | 16 Aug 2022 09:09 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/28357 |
Actions (login required)
 |
View Item |