ELEC2302 Project Ideas
Embedded Machine Learning & Edge AI Applications
Spasticity Detection using EMG Signals
Deploy lightweight machine learning models on embedded hardware to analyze electromyography (EMG) sensor data, detecting muscle activation patterns indicative of spasticity for real-time neurological monitoring.
Context-Aware Voice Transcription with LLM
Build an intelligent medical transcription system for patient notes that leverages free LLM models via OpenRouter API to perform context-aware correction of speech-to-text errors, improving clinical documentation accuracy.
Embedded Neural Architecture Search (eNAS)
Investigate automated neural architecture search techniques specifically designed for TinyML environments, optimizing compact models for ultra-low-power embedded platforms with strict memory and computational constraints.
SensorLLM: Edge Sensor-to-Language
Create a lightweight interface that transforms raw sensor data into natural language summaries at the edge, enabling human-readable insights from IoT sensors without cloud dependency for real-time environmental understanding.
Keyword Spotting Emergency Beacon
Implement a voice-triggered emergency alert system using on-device keyword detection with microphone sensors, enabling hands-free distress signaling for safety-critical applications and elderly care. Look into reverse geo-coding using scanned access point names, also through the IP. Combine this with GPS location for improved indoor accuracy.
AgriAgents: Smart Crop & Soil Analysis
Deploy image-based TinyML models on embedded vision boards to classify plant health conditions and assess soil quality directly at the field edge, enabling precision agriculture without internet connectivity.
Object Classifier & Counter
Build a computer vision system on Xiao microcontroller to count and track objects in real-time (people, vehicles, animals), with ML-based relationship detection to identify patterns like same-person entry/exit events.
Tooth Decay Detection at Edge
Develop an on-device dental health classifier using a curated tooth image dataset, enabling embedded systems to assess cavity severity for point-of-care diagnostics and teledentistry applications.
Geofencing & IoT Tracking
Create an intelligent geo-fencing and asset tracking solution for warehouse and supply chain logistics, utilizing low-power IoT devices for real-time location monitoring and boundary violation alerts.
Radar & mmWave Vision Fusion
Implement multimodal sensor fusion combining radar and camera data for robust object detection and tracking, processing both modalities locally using TinyML for enhanced perception in all conditions.
TinyML for Autonomous Vehicles
Deploy low-resolution sensor processing on microcontrollers for basic obstacle recognition and lane detection, demonstrating foundational autonomous driving capabilities on resource-constrained hardware.
PPG-Based Blood Sugar Estimation
Develop lightweight regression models to estimate blood glucose trends from photoplethysmography (PPG) signals, enabling non-invasive continuous monitoring on wearable devices.
Smart Bandages with TinyML
Create intelligent wound care monitoring by classifying healing progress based on temperature and color changes from sensor data, enabling proactive healthcare intervention detection.
Synthetic Data for TinyML
Explore generative approaches to augment limited real-world datasets for training compact embedded models, investigating GAN and diffusion techniques for improving TinyML model performance.
TinyML with PoseNet & SMPL
Implement efficient posture recognition and motion classification at the edge using optimized pose estimation networks for applications in fitness tracking and physical therapy monitoring. Combine poseNet with SMPL.
Asthma Detection System
Analyze breathing patterns and cough sounds to detect early signs of asthma exacerbation on embedded platforms, providing timely alerts for respiratory health management.
Smart Blind Stick with TinyML
Integrate radar or ultrasonic sensing with TinyML classification to detect and alert visually impaired users about obstacles, enhancing mobility independence and safety.
Tiny BERT for MCUs
Implement a compressed bidirectional encoder for on-device natural language understanding and command recognition, bringing transformer capabilities to microcontrollers.
Reinforcement Learning at MCU
Apply lightweight reinforcement learning algorithms (Q-learning, policy gradients) for adaptive embedded control tasks, enabling devices to learn optimal behaviors on-device.
Livestock Monitoring System
Use accelerometer and audio data to detect animal movement, feeding patterns, and distress signals at the edge for precision livestock farming and animal welfare monitoring.
Indoor Hazard Detection
Combine environmental sensor readings (COâ‚‚, temperature, humidity) with local TinyML models to detect unsafe indoor conditions and trigger automated alerts for building safety.
Concealed Metal Detection with Radar
Use radar signal processing and embedded TinyML classification to identify concealed metallic objects for security screening applications on low-power devices.
GI Tract Monitoring
Analyze motion and bio-signal data to identify abnormal gastrointestinal tract activity through on-device inference for digestive health monitoring and early disease detection.
TinyML Robot Navigation
Train on-device models to classify navigation cues distinguishing obstacles from clear paths, enabling autonomous movement for resource-constrained robotic systems.
TinyML for Biodiversity
Deploy environmental audio analysis for species recognition from bird calls and insect sounds using TinyML, supporting wildlife conservation and ecosystem monitoring.
Capacitive Leak Detection
Process capacitive sensor readings locally to detect fluid leaks or dielectric property changes for industrial monitoring and preventive maintenance applications.
Vibration-Based Communication
Explore vibration signal patterns as a novel communication or alert mechanism using local classification, enabling covert or alternative communication channels.
Forest Fire Vision Detection
Deploy on-board image classifiers for smoke and flame detection in outdoor forest environments, enabling early fire warning systems with edge processing.
Cough Recognition System
Implement embedded audio models to recognize cough sounds for health monitoring, infection detection, and respiratory disease screening in public health applications.
Anomaly Detection in Sensor Networks
Use autoencoder-based TinyML models to detect sensor anomalies or drift in distributed IoT networks, ensuring data integrity and system reliability.
Multimodal Vital Signs Fusion
Combine EMG, PPG, and accelerometer data to estimate fatigue, stress, and physical exertion locally, enabling comprehensive health monitoring on wearable devices.
Radar Object Proximity & Speed
Train TinyML models to estimate object speed and distance from radar reflections for edge-based sensing in automotive, robotics, and security applications.