University of Twente Student Theses
Evaluating pose estimation and object detection models for the application in the minisoccerbal project
Ahmadov, Parviz (2022) Evaluating pose estimation and object detection models for the application in the minisoccerbal project.
|
PDF
664kB |
| Abstract: | The focus of this research is to evaluate several lower extremity and sports ball detection models to determine whether the combination of the two can later be used to calculate several parameters of the exercises that are performed with a soccer training object called minisoccerbal. There are already a number of existing machine learning libraries and algorithms that are able to detect lower body joints and a sports ball in an image. This research aims to examine existing tools’ effectiveness in detecting the required objects in videos that are within constraints relevant to the MiniSoccerball project, that is, the use of a mini soccer ball, stationary camera, etc. Aside from the accuracy of algorithms, processing speed is also a priority, with real-time object detection being the future direction. Two object detection models (for minisoccerbal detection) - YOLOv5 and EfficientDet, and two pose estimation models - OpenPose and BlazePose (for lower body joint detection) were chosen for evaluation. mAp (Mean Average Precision) scores were used for evaluating object detection models while Pose estimation models were evaluated based on PDJ scores (Percentage of Detected Joints). FPS (Frame per second) was calculated for determining the processing speed for all the models. Although EfficientDet had a slightly higher mAP score compared to YOLOv5, YOLOv5 was chosen as the more suitable model because of the speed advantage and having a sufficiently high mAp@0.75 score. For pose estimation, OpenPose was determined to be more suitable despite being significantly slower, due to BlazePose having a lower PDJ score. |
| Item Type: | Essay (Bachelor) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Computer Science BSc (56964) |
| Link to this item: | https://purl.utwente.nl/essays/91859 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page