Perception tools and their usage in sports
Kilpeläinen, Roni (2023)
Kilpeläinen, Roni
2023
Tekniikan ja luonnontieteiden kandidaattiohjelma - Bachelor's Programme in Engineering and Natural Sciences
Tekniikan ja luonnontieteiden tiedekunta - Faculty of Engineering and Natural Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2023-06-09
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202306016403
https://urn.fi/URN:NBN:fi:tuni-202306016403
Tiivistelmä
Computer vision and perception tools are an important step on the developing world of robotics. The goal of computer vision development is to enable machines to understand visual information. When this goal is acquired, there are countless usages in fields such as production, healthcare, and entertainment like sports. This thesis addresses perception tools, computer vision and especially their usage amongst sports.
The thesis can be roughly divided into two parts, the first part is literature research about computer vision and perception tools. The second part of the thesis is a practical example with OpenDR’s pose estimator tool, which is used to do a short analysis on two separate golf swings. These swings are then compared to each other and by this one way of using perception tools in sports is showcased.
First part of the literature research addresses the basics of computer vision. The beginning of this part introduces what computer vision and perception tools are. In addition, a rough version of a vision system is presented. After that we move on to how computers process images. Briefly computers segment the picture and then give a value to each part that corresponds into a color. After image processing we switch over to 3D-vision and present three different ways on how it is executed. The methods that were presented are, triangulation, time-of-flight, and structured light. Lastly, we researched on how object tracking is implemented on vision systems. We found out that no one way of executing object tracking is set in stone, and there are available several different algorithms. One common factor in these algorithms was the need of background filtering. Usually this is done by comparing subsequent images and separating objects that are in motion from those that are motionless. At the end of this part project OpenDR is also showcased, which is used in the practical part of the thesis.
The second part of the literature research contains presentations of different systems that are already used in sports. In total 4 systems were presented, each of them focusing on a different part of computer vision. The presented systems were used in hockey, tennis, cricket, and badminton. The purpose of these systems include improvement in training efficiency, added entertainment value, and increasing the accuracy of officiating. Some of the systems are already in commercial use and some were still in development. Although the purpose of these systems is to develop sports in general, there are some difficulties. These difficulties include technical ones like accuracy of the system or price, and impact on the entertainment value by disrupting the flow of the game.
On the last part of the thesis a pose estimator algorithm was used to detect joints and other keypoints from images. These keypoints were used to compare two different golf swings and to detect differences in these swings. These differences were then used to consider on what should be changed on the swing to acquire better results. The algorithm made it easier to detect the differences in the swings and we concluded that with more data such as different angles, in addition to better sport knowledge, more accurate advice could be provided.
The thesis can be roughly divided into two parts, the first part is literature research about computer vision and perception tools. The second part of the thesis is a practical example with OpenDR’s pose estimator tool, which is used to do a short analysis on two separate golf swings. These swings are then compared to each other and by this one way of using perception tools in sports is showcased.
First part of the literature research addresses the basics of computer vision. The beginning of this part introduces what computer vision and perception tools are. In addition, a rough version of a vision system is presented. After that we move on to how computers process images. Briefly computers segment the picture and then give a value to each part that corresponds into a color. After image processing we switch over to 3D-vision and present three different ways on how it is executed. The methods that were presented are, triangulation, time-of-flight, and structured light. Lastly, we researched on how object tracking is implemented on vision systems. We found out that no one way of executing object tracking is set in stone, and there are available several different algorithms. One common factor in these algorithms was the need of background filtering. Usually this is done by comparing subsequent images and separating objects that are in motion from those that are motionless. At the end of this part project OpenDR is also showcased, which is used in the practical part of the thesis.
The second part of the literature research contains presentations of different systems that are already used in sports. In total 4 systems were presented, each of them focusing on a different part of computer vision. The presented systems were used in hockey, tennis, cricket, and badminton. The purpose of these systems include improvement in training efficiency, added entertainment value, and increasing the accuracy of officiating. Some of the systems are already in commercial use and some were still in development. Although the purpose of these systems is to develop sports in general, there are some difficulties. These difficulties include technical ones like accuracy of the system or price, and impact on the entertainment value by disrupting the flow of the game.
On the last part of the thesis a pose estimator algorithm was used to detect joints and other keypoints from images. These keypoints were used to compare two different golf swings and to detect differences in these swings. These differences were then used to consider on what should be changed on the swing to acquire better results. The algorithm made it easier to detect the differences in the swings and we concluded that with more data such as different angles, in addition to better sport knowledge, more accurate advice could be provided.
Kokoelmat
- Kandidaatintutkielmat [8798]