Stationary LiDAR for Traffic and Safety Applications \u2013 Vehicles Interpretation and Tracking

015446202014PDFTech ReportNEXTRANS Project No. 134PY2.1DTRT07-G-005Laser radarRadar vehicle detectorsVehicle detectorsRadar trackingAutomatic vehicle detection and identification systemsData collectionData storageFeasibility analysisIntersectionsSensorsTraffic dataTurning trafficLiDART-ScanData and Information TechnologyHighwaysOperations and Traffic ManagementI72: Traffic and Transport PlanningIndianaUnited StatesNextransTarko, Andrew P.Ariyur, Kartik B.Romero, Mario A.Bandaru, Vamsi KrishnaLiu, ChengPurdue UniversityNextransUS Transportation CollectionThe goal of the T-Scan project is to develop a data processing module for a novel LiDAR-based traffic scanner to collect highly accurate microscopic traffic data at road intersections. T-Scan uses Light Detection and Ranging (LiDAR) technology that can detect and track various types of road users, including buses, cars, pedestrians, and bicycles; and, unlike video detection, it does not experience the well-known occlusion problem. Moreover, LiDAR data has a one-to-one correspondence with the physical world, which makes it possible in principle to produce the positions and velocities of road users in real-time as needed for traffic and safety applications, with the errors of estimation dependent only on the resolution and accuracy of the LiDAR sensor. We faced two major challenges to this goal after integration of the sensing, data collection, and processing components: 1) the reduction of resolution farther from the sensor because of fewer light rays per unit area and reflections that do not return to the sensor and 1) the-wind induced oscillations of the sensor position. We solved the latter problem through installation of an inertial sensor atop the LiDAR to calibrate its motion. We are developing several approaches to solve the first problem, including video integration. A lot of our code is already open-source-based, and we have found from the computational requirements for various algorithms that the use of a standard GPU for processing should permit real-time running of our algorithms in the data processing module. System Integration: The system we integrated consists of a LiDAR sensor installed on a pneumatic 42-foot telescoping mast. The sensor head rotates 900 times per minute, which results in 1.3 million data points per second. Data collected over a period of several hours to several days is stored in high-capacity devices. The final stage of integration and hardware testing was performed as a part of this project. A customized aluminum base physically integrates the mast, pan/tilt mechanism, LiDAR head, and two video surveillance cameras to provide a human-friendly data format for supervision and evaluation. The T-Scan collects 3D data with 64 laser sensors arranged vertically in a fan-like manner. The spherical data (range, vertical angle, and horizontal angle) are first converted to more convenient XYZ Cartesian data. The XYZ data points are then interpreted as moving objects and the background.75