Introduction

Roadside traffic monitoring is used in many cities to monitor the usage and condition of roadways. This can be performed using a range of sensors; in this project we will build a prototype acoustic traffic monitoring system designed for low-power operation, which will detect and distinguish between cars and commercial vehicles. we will design and deploy our application using an ultra-low-power audio processing device Xylo Audio from SynSense.

The system will be designed using the acoustic-based traffic monitoring dataset and augmentation system from DCASE2024. When a vehicle is detected, the system would provide a sparse positive output indicating either a car or commercial vehicle. The system would provide no output at all other times. It is not necessary to use the multi microphone array data, but instead we may use monaural data. Likewise it is not necessary to detect the direction of travel; simply detecting and classifying the presence of vehicles is sufficient.

we will implement a live roadside demo, based on a laptop, as the final phase of the project. This will include a python notebook or web API front-end.

Xylo is a new audio inference processor, which encodes audio as sparse temporal events, and performs ML inference using networks of low-bit-depth, temporally sparse neurons (sometimes known as Spiking Neural Networks). In this project we will train an application to classify real-time audio input using Xylo, and deploy it to prototype hardware. Xylo supports simple neural network architectures (dense feed-forward; recurrent; residual).

Some Constraints and considerations

Here are some project constraints:

  1. The system’s functionality will be demonstrated in a roadside setting using the Xylo Development Kit connected to a laptop

  2. The vehicle classification model must be a Spiking Neural Network (SNN).

  3. The Xylo Development Kit must be the hardware platform for the system.

  4. The system must use an acoustic-based traffic monitoring system to classify vehicles.

Technical Part

The process of the research and development