Persistent Homology for Detection of Objects from Mobile LiDAR Point Cloud Data in Autonomous Vehicles

Abstract

Recently, researchers have paid significant attention to problems related to object detection and computer vision for autonomous vehicles. Such vehicles offer many benefits, including their ability to help address transportation-related issues such as safety concerns, traffic jams, and overall mobility. Multi-beam ‘light detection and ranging’ (LiDAR) is one of the main sensors that is used to sense and detect objects by creating a point cloud data map of the surrounding environment. Current object detection tasks that use only mobile LiDAR data divide the entire area into cubes and employ image object detection methods. Such an approach poses challenges due to the third dimension that increases computational time, which thus requires a tradeoff between performance and time optimization. In this paper, we propose a new approach to detect objects using point cloud data by investigating the shapes of the objects. To this end, we developed a method based on topological data analysis achieved via persistent homology to analyze the qualitative properties of the data. To the best of our knowledge, our work is the first to develop topological data analysis for real-world mobile LiDAR point cloud data exploration. The evaluation result shows a high accuracy classification result using features extracted from barcodes.