Herrmann, Nikolai (2022) Object Anchoring for Human-Robot Interaction: Connecting Sensor Data to Symbols. Bachelor's Thesis, Artificial Intelligence.
|
Text
bAI_2022_HerrmannN.pdf Download (5MB) | Preview |
|
![]() |
Text
Toestemming.pdf Restricted to Registered users only Download (120kB) |
Abstract
Tracking algorithms, which track objects in motion, tend to fail when objects get close to one another, overlap, or are temporarily occluded. To combat this limitation, the anchoring framework can be applied, allowing objects to be tracked symbolically such that they can be uniquely identified at any moment. This is achieved by maintaining correspondence between raw perceptual data from sensors and abstract symbols, using a matching function. Here, we use a bottom-up approach where the matching function either acquires a new object or reacquires a previously seen one. Four different binary classifiers were trained to accomplish this task. To avoid manual labelling we track objects separately by color to maintain a ground truth. In addition, two approaches of the reacquire functionality are explored: one which continuously anchors and one which anchors only stationary objects while still observing moving ones. Scenarios of different difficulties and category were tested including the Three-Card Monte and the Shell Game. Our results demonstrate limited success with the first approach due to data sensitivity, but the second approach shows clear improvements with the help of motion analysis and the Kalman filter.
Item Type: | Thesis (Bachelor's Thesis) |
---|---|
Supervisor name: | Mohades Kasaei, S.H. |
Degree programme: | Artificial Intelligence |
Thesis type: | Bachelor's Thesis |
Language: | English |
Date Deposited: | 28 Jul 2022 13:44 |
Last Modified: | 28 Jul 2022 13:44 |
URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/28199 |
Actions (login required)
![]() |
View Item |