Edge AI for Real-Time Object Detection on Low-Power IoT Devices

This research project evaluates the deployment of deep-learning-based object detectors on resource-constrained edge devices (such as a Raspberry Pi with a Coral/Hailo accelerator or an ESP32-CAM), where memory, compute, and power budgets are far smaller than typical cloud GPU environments.

Several lightweight architectures (MobileNet-SSD, YOLO-Nano, and a quantized YOLOv8n) are benchmarked after post-training quantization (INT8) and pruning, measuring inference latency, frames-per-second, model size, and mean average precision (mAP) trade-offs against the full-precision baseline.

The study also profiles power consumption during continuous inference to estimate battery life for a wearable/portable use case, and discusses practical techniques — such as frame-skipping and region-of-interest cropping — that further reduce compute load with only a modest accuracy penalty, informing real-world edge-AI deployment decisions.

Detailed Project Description

Edge AI for Real-Time Object Detection on Low-Power IoT Devices applies computer vision to camera frames to detect or recognise what is in view and act on the result automatically.

How It Works
1. The camera captures frames continuously.
2. Frames are pre-processed and fed to the detection/recognition model.
3. Detections are drawn on the output and used to trigger the required action (log, alert, unlock, count).
4. Results and images are stored or streamed to a browser/dashboard.

Key Components and Their Role
- ESP32-CAM: ESP32 module with an OV2640 camera used to capture images or stream video.
- Raspberry Pi: single-board Linux computer that runs the higher-level processing and networking.
- Camera module: image sensor used for capture, streaming or computer vision.
- Li-ion battery + charger (TP4056): portable power with charge management.
- Edge Impulse: TinyML platform used to train and deploy the on-device model.
- TensorFlow: machine-learning framework used to build and train the model.
- YOLO: real-time object-detection model family.

Implementation Steps
1. Collect the components and study each module's datasheet (pin-out, voltage, interface).
2. Wire the circuit on a breadboard following the pin mapping; power sensors from the correct rail (3.3 V / 5 V).
3. Install Python, TensorFlow Lite, Edge Impulse and the required libraries; flash a minimal test sketch for each module separately.
4. Write the main firmware: read sensors → process → drive outputs / send data.
5. Configure network or cloud credentials and test communication end-to-end.
6. Calibrate thresholds, add error handling (reconnect, sensor time-out) and run a long-duration test.
7. Finalise on a PCB / enclosure and document wiring, code and results.

Key Features
- Real-time detection
- Annotated output
- Event logging
- Adjustable confidence threshold

Expected Output
On completion the student will have a live video/image feed with detected objects highlighted and the corresponding actions logged.

Deliverables
- Complete source code with comments
- Circuit diagram and pin mapping
- Project report with screenshots and results
- Demonstration-ready build