Developing a Smart Child Presence Detection in a Vehicles Using Thermal Camera

Abstract

Children left unattended in parked cars face severe risks due to rapidly rising cabin temperatures, often leading to heatstroke or fatalities. Even on moderately warm days, interiors can reach life-threatening levels within minutes, and despite awareness campaigns, these incidents continue, highlighting the urgent need for proactive detection systems.This project presents a smart child presence detection system for vehicles using thermal imaging and deep learning, designed on preventing fatalities caused by vehicular heatstroke. The system integrates the MLX90640 thermal camera with a Raspberry Pi 5 and leverages the YOLOv8s object detection model to identify human heat signatures in real time. Unlike conventional camera-based or multi-sensor solutions, this work introduces a lightweight, cost-effective approach that ensures reliable detection regardless of lighting conditions. A key novelty of this system lies in its fully embedded implementation, performing all processing locally on the Raspberry Pi to enable fast, autonomous operation without external infrastructure. The system is designed to trigger an alert when a human is detected in the vehicle cabin, activating a buzzer and sending a thermal image notification to the caregiver via a Telegram bot. A five-minute delay mechanism is included to prevent repeated alerts unless manually reset. The YOLOv8s model was trained on a custom thermal dataset and tested under varying distances, angles, and body positions to simulate realistic in-vehicle conditions. Comparative evaluation with a commercial infrared thermometer confirmed the MLX90640’s capability in detecting human heat signatures. The results demonstrate the feasibility and potential of combining thermal sensing with on-device AI for real-time child presence detection in vehicles.