Interest in GSoC 2025 project:Project: Embedded AI Accelerator for OpenCV on Microcontrollers

[Devika A]

I hope you are doing well.
My name is Devika.A, a Second Year Applied Electronics and Instrumentation student from India
Project: Embedded AI Accelerator for OpenCV on Microcontrollers Overview:

Integrate lightweight AI models into OpenCV’s DNN module for microcontrollers, optimizing inference speed and memory efficiency. This would help run AI-based vision applications on low-power devices like ESP32, STM32, or Raspberry Pi.

Approach:

1. Optimized Model Deployment: Convert pre-trained deep learning models (like YOLO, MobileNet) into formats optimized for microcontrollers (e.g., TensorFlow Lite, ONNX, or Edge Impulse).

2. Efficient Computation: Use integer quantization (INT8) or half-precision floating point (FP16) to reduce the model size and improve execution speed.

3. Hardware Acceleration: Utilize microcontroller-specific optimizations like CMSIS-NN for ARM Cortex chips or ESP-DSP for ESP32.

4. Benchmarking: Develop a test suite to evaluate performance, memory usage, and inference time on different embedded platforms.

It can be a great portfolio project for future job opportunities in robotics or AI on edge devices.