AN AIRBORNE LIDAR SCANNING AND DEEP LEARNING SYSTEM FOR REAL-TIME EVENT EXTRACTION AND CONTROL POLICIES IN URBAN TRANSPORTATION NETWORKS

Abstract

Airborne light detection and ranging (lidar) and unmanned aircraft systems (UAS), also called drones, are emerging technologies that are of growing interest for a range of transportation applications. At the same time, machine learning is leading to rapid advances in the ability to automatically extract actionable information from lidar data. This project investigated the combined use of UAS, lidar and machine learning for traffic network monitoring. A custom lidar-UAS was built, instrumented, and used to acquire data over multiple test sites. New processing algorithms were developed to automatically parse raw data, generate georeferenced point clouds, filter out repetitions, and perform scanning to identify vehicles, all with processors that can be mounted on UAS and operated in real-time. The results of the UAS lidar data collection were used to develop operational guidance for transportation agencies using UAS to collect lidar data in proximity to active roadways. Additionally, this work resulted in an end-to-end processing system, implemented in C++, capable of real-time vehicle recognition with processors that can be mounted on UAS. A final output of the project was a set of specific recommendations for follow-up research, contributing to the long-range vision for traffic networking monitoring using a fleet of UAS that can inform real-time decisions.