Improving distance and impact location measurements in athletic sports using state-of-the-art machine learning/machine vision.
This thesis project is an excellent way to get hands-on experience using AI to solve complex computer vision problems and learn how to go from a system design to a hardware prototype. This thesis-project aims to design and build a prototype of a camera-tracking system that can determine where objects land within a fixed area. Although similar systems already exist today, the novelty of this project lies in the fact that we are trying to increase the area that can be monitored.
Background: Did you win the Olympics, or did you “only” finish fourth?
This question is a reality for many athletes in today’s world, especially in events such as javelin throw, discus throw, hammer throw and, shot put. To measure where the object lands is a manual task conducted by humans; therefore, mistakes and errors in the measurement process are not uncommon. The person making the measurements must avoid being hit by a deadly projectile, which does not make the task any easier. Imagine training your entire life and not getting that well-deserved medal because of a measurement error. To prevent this from happening, we want to try and design a computer-vision system that can assist in the measurement process for the events mentioned above. Since tracking flying objects is a general task, a functional prototype could be of interest in many other areas and not just sports events. You do not need to be interested in these events or have any previous knowledge of the events to complete this project.
Problem formulation: Can we use modern and intelligent technology (AI) to measure the length of the throws of one or several of the following events: javelin throw, discus throw, hammer throw, and shot put, or should it only be used as a complement to the current measurement process.
Target: To construct a prototype using a set of cameras connected to a computer that can measure the length of the throws, we want the system to present a result in under 10 seconds from the moment of impact.
- Below is a compiled list of some of the questions/problems that need to be addressed in this project.
- How do different surfaces .e.g grass, gravel, plastic, affect the system?
- How do we make sure that the system only records the first impact (objects can bounce)?
- How to make correct measurements when the object lands abnormally, .e.g a javelin that lands flat on the ground?
- How many camera sensors are needed, and where should they be placed relative to the thrower?
- Which type of camera sensors are best suited for this task?
- Which AI model should be used to achieve a good mix of accuracy and performance?
- How do we collect and store data that the model can be trained on?
Required background: Image processing, AI/ML, CNN, OpenCV, Python, or C++.
Start date: January 2022. End date: June 2022.
Contact: Niclas Jansson – BitSim Now