Mentor Visit and Product Progress!

During my mentor visit, I discussed with my mentor about how I could start programming my project given my research and study on convolutional neural networks, machine learning tools such as MediaPipe and OpenCV, and hand gesture projects on sign language and detection. We mainly discussed on the approach for the detection of the palm/hand and how we can create a bounding box to analyze a specific aspect of a photo frame. Our 3 main techniques that we want to try during this project are motion detection (using different frames of a video to figure out a hand's position), skin detection (using a specific value of skin color and essentially finding pixels that have the most similar values as the skin color), and background subtraction (taking the value of the contrast between the background and the skin to figure out the position of the hand). We also discussed how there is a lot of trial and error required to get a sophisticated model for this project. Another thing that I learned about was Transfer learning which basically means to use a pre-existing model that is well-trained with thousands of data (images in this case), but build on its layers with your own different layers and data to create a model with a better precision at classifying images. Lastly, my main objective is to reduce my data to only 4-5 specific hand gesture classes and take 60-70 images on them, program the hand detection of my model, and decide on whether preprocessing is required on my data.

Currently, I am planning to finish gathering up my data, making key decisions on preprocessing and applications of my project, and starting my programming for hand detection to ultimately create a convolutional neural network that can identify hand gestures successfully.

So see you next week, same place, same time.