Embedded Pothole Detection and Mitigation System using NAS-Optimized CNN and IoT

Potholes in roads are a dangerous menace that causes accidents, damages vehicles, and incurs expensive repair bills. This paper introduces an affordable, real-time pothole detection and classification system based on embedded systems and intelligent image processing. The solution proposed uses a NAS-optimized Convolutional Neural Network (CNN) with ultrasonic sensing to improve detection accuracy while minimizing false positives. The system is deployed on a robot platform with an ESP8266 microcontroller, ultrasonic sensor, L298N motor driver, and Blynk IoT for remote monitoring. Real-time image data is processed by the NAS-CNN model, and validation of distance is done using ultrasonic sensors to make the detection robust. Notifications are sent through the Blynk IoT platform, and an auto speed control mechanism reduces collision risks. The model was learned from more than 2,000 labeled road images acquired from RoboFlow and GitHub. Experimental results show a 97% validation accuracy with a 15% decrease in false positives over traditional CNNs. The system shows near-zero latency in actual usage in field tests, affirming it to be ready for embedded deployment. This prototype exhibits high detection capability, scalability for use in smart vehicles, and proactive road maintenance as part of a smart city infrastructure for enhancing road safety and urban mobility.