Currently, most computer vision algorithms are focused on a frame by frame RGB data analysis. In this experiment, students explored the use of infrared depth data from a Microsoft Kinect as a tool to classify a set of full-body exercises. The model was then optimized using two different approaches, brute force and data driven. The approaches were graded on their ability to increase the accuracy of the model. Using this information, future researchers will be able to develop models that can efficiently and effectively perform task detection and segmentation of similar datasets.

Throughout the course of the 10 weeks I worked on the project I gained valuable experience navigating around issues in Machine Learning, Big Data, and Project Management. Through my interactions with my coworkers and faculty advisors I gained valuable insight into industrial and academic practises, problem solving approaches, and communication skills.

If have any questions feel free to reach out to me!

Download Poster PDF See on Github