Klasifikasi Kendaraan Berbasis Suara Di Lalu Lintas: Implementasi TinyML

Abstract

Driving is deeply embedded in Indonesian culture. All Indonesian citizens (WNI) who meet the qualifications can obtain a driver's license (SIM). As a result, the number of drivers in Indonesia is relatively significant, resulting in a high rate of traffic accidents. One of the reasons leading to the rising number of road accidents is hearing loss. This study focuses on building a tool that can classify vehicles based on the audio of the vehicle while it is running in order to assist drivers with hearing impairments. This study employs the Arduino Nano 33 BLE Sense, which includes an embedded microphone, and the Edge Impulse open platform. A simple artificial neural network model with four 1-D convolution layers and one hidden layer was utilized. The dataset was obtained manually using the YouTube open platform and was captured directly on the Edge Impulse platform using the Arduino Nano's microphone. The model training procedure was repeated 80 epochs. During training and testing, the model is regarded to be accurate, with accuracy scores of 96.3% and 90.9%, respectively. With these adequate results, the model is judged eligible for implementation, despite the fact that it still requires further refinement.