Road accidents remain a leading cause of global fatalities, with over 1.35 million deaths annually according to WHO. Existing vehicle safety systems rely on passive mechanisms—airbags, ABS, ESC—that mitigate injury severity but cannot predict or prevent accidents, nor communicate emergencies automatically. This paper presents a Smart Vehicle Safety System that integrates Internet of Things (IoT) sensing with embedded Machine Learning (TinyML) to address these limitations. The system employs an ESP32 microcontroller interfaced with an MPU6050 accelerometer, HC-SR04 ultrasonic sensor, NEO-6M GPS module, SIM800L GSM module, NRF24L01 wireless transceiver, and a DC motor with PWM driver. A Random Forest classifier, trained on accelerometer data and compressed into a C++ model header via Eloquent ML, performs real-time on-device accident detection without cloud dependency. Upon detection, the GPS coordinates are transmitted as SMS alerts to emergency contacts through the GSM module, while accident records—including timestamp, location, and impact intensity—are uploaded to a Firebase Realtime Database for remote monitoring. A zone-based adaptive cruise control feature uses RF signals from roadside NRF transmitters to automatically regulate vehicle speed in critical zones such as school areas, cross-roads, and city limits. Experimental validation demonstrated 94.3% accident detection accuracy across 35 simulated impact events, emergency SMS delivery within 8–12 seconds, and reliable zone-based speed regulation with sub-200 ms response. The system provides a cost-effective, scalable solution for next-generation intelligent transportation systems.

Keywords—IoT; TinyML; ESP32; Accident Detection; Adaptive Cruise Control; GPS; GSM; Firebase; NRF24L01; Random Forest; Vehicle Safety System