Free Project - Computer Vision and Embedded Systems

This repo has a computer vision project with CNN object detection, an embedded kernel for ESP32, and a synthetic data generator.

What it does

• Detects obstacles and large objects in front of the user
• Covers a 120° field of view
• Works at 15-30m range with ~50-100 MOA resolution
• Gives real-time haptic feedback based on 3D awareness
• Low latency (<100ms) and runs at 10-15 FPS
• Battery lasts all day, ideally a full work week

Project structure

free-projet/
├── cnn/                    # CNN object detection
│   └── proto_v1/          # First prototype with PyTorch
│       ├── model.py       # ResNet18 detection model
│       ├── train.py       # Training script
│       ├── main.py        # Inference and visualization
│       ├── data/          # Data loading and augmentation
│       └── README.md      # CNN docs
├── kernel/                # ESP32 kernel
│   ├── src/               # Kernel source
│   ├── include/           # Headers
│   └── Makefile          # Build system
└── synthetic_generator/   # Zig synthetic data generator
    ├── src/               # Generator source
    ├── build.zig          # Build configuration
    └── README.md          # Generator docs

Quick start

CNN Object Detection

cd cnn/proto_v1
make activate          # Activate virtual environment
make train             # Train the model
make inference         # Run inference on sample image

Embedded Kernel

cd kernel
make all               # Build kernel
make flash             # Flash to ESP32 (needs esptool)

Synthetic Data Generator

cd synthetic_generator
zig build run          # Generate synthetic dataset
make test              # Run tests

Requirements

• CNN: Python 3.8+, PyTorch, CUDA (optional)
• Kernel: ESP32 toolchain, esptool.py
• Generator: Zig compiler (0.15.0+)

Docs

• CNN Documentation - Object detection guide
• Kernel Documentation - Embedded system stuff
• Generator Documentation - Synthetic data generation

License

See LICENSE file.