Week 2: Attempts to Train the AI

Weekly Task Overview

This week, we attempted to create a training dataset using images of rock, paper, and scissors gestures captured by our group members. We used this dataset to train an AI model and implemented it for real-time recognition and detection using a Raspberry Pi camera.

What we have achieved-Week 2

To train the AI to accurately recognize rock, paper, and scissors hand gestures, we need to create a comprehensive training set comprising hand images sourced from both the internet and group members. These images should encompass a variety of poses, angles, and skin tones to ensure diversity and accuracy.

   

The next step involves manually tagging and classifying the images into three categories based on the hand gestures. This process requires carefully selecting and annotating the regions containing relevant information. Once a sufficient number of training samples have been prepared, the AI can be trained autonomously.

Upon completing the training, we will obtain the AI's performance results, providing a summary of how effectively it has learned to recognize each gesture. The training dataset is initially stored in a zip file, requiring specific commands to convert it into a format compatible with the Raspberry Pi camera.

After launching the program, the camera will be able to identify hand gestures in real time. This newly converted file allows for seamless recognition, enabling the camera to instantly detect gestures upon execution.

Raspberry Pi Real-Time Detection Screenshot

After running the program, it was observed that the device can accurately detect hand gestures in real-time under simple background conditions. The detected gestures are highlighted with bounding boxes, along with the corresponding confidence levels displayed on the screen.

To further improve the accuracy and stability of real-time detection, our group expanded the training dataset and fine-tuned the program code. However, it is important to note that the Raspberry Pi camera should avoid direct light sources in the background, as it does not automatically adjust exposure. Intense light can interfere with the camera's ability to accurately recognize images.