Real Time Hand Gesture-Based Light Control System Using Computer Vision and Machine Learning

In this paper, we aim to develop a full-stack system to control LEDs based on hand gestures using computer vision and machine learning methods. The system recognizes finger movements using the MediaPipe library, classifies the number of raised fingers based on the KNN (K-Nearest Neighbors) and transmits the result information to the Arduino board for the use of the information. A database of finger poses taken in different light and filming angles was collected, this data set was randomly split into 80% train and 20% test. Different values of the number of neighbors (k) have been tested, and the accuracy obtained is 92% for k = 5. The operation had more efficiency in larger hands with the mean response time of 0.15 seconds, which was acceptable for real-time interaction.