Ai-driven Edge-based Smart Parking System With Nfc Authentication For Parking Slot Recommendation And Access Time Prediction

Abstract

Efficient Parking Management Requires Intelligent, Real-time Decision-making To Minimize Search Time And Improve Space Utilization. This Paper Presents An AI-driven, Edge-based Smart Parking System Built Around A Comparative Evaluation Of Random Forest And Neural Network Models For Parking Slot Recommendation And Parking Access-time Prediction, In Which The Neural Network Model Was Selected For Its Higher Accuracy, Consistency, And Smaller Model Size. The Selected Neural Network Models Were Independently Trained As Two TinyML Regression Models In Edge Impulse (test_slot And Test_time), Deployed In Float32 Form, And Merged Into A Single Inference Library (test7time_inferencing_v2.h) Running On An ESP32-WROOM-DA. The Implemented System Performs NFC-based User Authentication Using A PN532 Reader/writer, IR-sensor-based Parking-slot Status Detection Across A 3-lane, 9-slot Layout, Lane-wise Congestion Calculation, On-device Slot Recommendation Based On The Lowest Suitability Score, User Confirmation Through A 2.8-inch SPI TFT Touch Display, On-device Access-time Prediction, Servo-actuated Barrier Control, And RTC DS1307 With Logger.py-based Real-hardware Data Logging. Both TinyML Models Achieved Strong Test-set Performance, With 97.26% Accuracy For Slot Recommendation And 98.68% For Access-time Prediction, While Maintaining Approximately 3 Ms Inference Latency And RAM Usage Below 1.5 KB On The ESP32. Real-time Behavioural Testing Confirmed That The Deployed Models Respond Correctly To Changes In Slot Status, Congestion, And Distance, Dynamically Generating Slot Scores And Access-time Predictions, And Real-hardware Testing Confirmed Successful Integration Of NFC Authentication, IR-based Slot Sensing, TinyML Inference, User Interaction, Barrier Control, RTC Timestamping, And CSV-based Data Logging. The Results Demonstrate That The Selected Neural Network Models Can Be Deployed And Operated As A Real-time, Authenticated, Edge-based Smart Parking System Without Dependence On Cloud-based Processing.

Keywords