Gesture Neural Network (GNN) Part 1
Tuesday, Feb 17, 2026 | 2 minutes read | Update at Wednesday, Feb 18, 2026
A neural network for categorising inputs from sensors from a glove or similar into text.
This project is based on Simple Neural Network in Python from Scratch by Pradyumna
Project Link : Github
So, I’ve had this idea for quite some time, essentially a glove or some other method where I can use hand gestures as inputs. The input generated would probably be an array of at least five inputs, one for each finger, and potentially a sixth gyro sensor reading. Now, I could have just used something like a 5D array that could map to each finger sensor, but that’s boring. I decided instead to create a neural network from scratch. I didn’t want to use PyTorch or somee other ML library, because I wantedd to actually understand how neural nets work.
My goals for this project are:
Base Goals
- [] Working Neural Network
- [] Scalable number of inputs and output categories
- [] To get a solid understanding of how machinee learning works
Additional milestones
- [] Fast and efficient training
- [] >99% accuracy after training
- [] Easy user calibration, possibly a GUI for end-users
1 : Actually getting the base Neural Net
The first problem is that I don’t know anything about how Neural Networks (NNs) work. So I spent some time researching the basic flow from input to output.
NNs consist of a bunch of nodes, called neurons, organised into layers. In a basic NN, these layers are the Input, Hidden and Output layers. Input layers hold the input features, usually a dataset. Hidden layers hold the artificial neurons, that process the input features into new representations, through a mix of weights and biases, and aree passed through to thee output layer. This layer applies non-linear activation functions, simulating neurons firing, to produce a final prediction.
After learning the basics, I wanted to try to actually seee if a neural network could work, and simple search led me to Simple Neural Network in Python from Scratch by Pradyumna. It’s a great article, and the NN worked perfectly after a few tweaks to get it to process my own dataset. The NN by Pradyumna already comes with a configurable number of neurons for each layer, ticking off the scalability goal.
