Event Details
Dynamic gesture classification using machine learning method
Presenter: Hao Li
Supervisor:
Date: Mon, April 24, 2023
Time: 10:00:00 - 11:00:00
Place: via Zoom - please see link below
ABSTRACT
Join Zoom Meeting
https://uvic.zoom.us/j/83301231332?pwd=OVdWT1cxUDdzbk14TlVLWUM5Vy9VZz09
Meeting ID: 833 0123 1332
Password: 585901
One tap mobile
+16475580588,,83301231332#,,,,0#,,585901# Canada
+17789072071,,83301231332#,,,,0#,,585901# Canada
Dial by your location
+1 647 558 0588 Canada
+1 778 907 2071 Canada
Meeting ID: 833 0123 1332
Password: 585901
Find your local number: https://uvic.zoom.us/u/kh4aOgjTF
ABSTRACT:
Dynamic hand gesture recognition has been applied to many areas such as virtual reality, video games, remote robot operation and hand injury recovery. The appearance of Leap Motion Controller accelerated the development of this area with its powerful hand tracking technique. It can gather hundreds of features of hands which make it catch details of hands with accuracy.
In this project, ten single-hand gestures are classified using four machine learning algorithms which include random forest, linear support vector machine, logistic regression as well as K-nearest-neighbor. The performance of each method on each gesture is evaluated and compared. Linear support vector machine and K-nearest-neighbor methods give the best performance with accuracy over 90% but random forest gives the worst result whose accuracy is 69%. The reason for the performance difference is also analyzed which is caused by the small size of the training dataset that causes overfitting. Principal component analysis is used to reduce the dimension of feature space to make data density larger. A better result of random forest is obtained which proves the analysis. A program which can classify the gesture in real time is implemented which can give the correct result.