A Sensing Methodology for an Intelligent Traffic Sign Inventory and Condition Assessment Using GPS/GIS, Computer Vision and Mobile LiDAR Technologies
MetadataShow full item record
Traffic signs, which transportation agencies must inventory and manage, are one of the most important roadway assets because they are used to ensure roadway safety and provide important travel guidance/information. Traffic sign inventory and condition assessment are two important components that are essential for establishing a cost-effective and sustainable traffic sign management system. Traditionally, state departments of transportation (DOTs) have conducted traffic sign inventory and condition assessment manually, a process that is labor-intensive, time-consuming, and sometimes hazardous to field engineers in the roadway environment. Methods have been developed to automate sign inventory and condition assessment using video log images in previous studies. However, the performance of these methods still needs to be improved. Based on the need to inventory signs and manage them more effectively, this study has two focuses. The first focus is to develop an enhanced traffic sign detection methodology to improve the productivity of an image-based sign inventory for state DOTs. The proposed methodology includes two enhanced algorithms: a) a lighting dependent statistical color model (LD-SCM)-based color segmentation algorithm that is robust to different image lighting conditions, especially adverse lighting and b) an ordinary/partial differential equation (ODE/PDE)-based shape detection algorithm that is immune to discontinuous sign boundaries in a cluttered background. The second focus of the study is to explore a new traffic sign retroreflectivity condition assessment methodology to develop a mobile method that uses emerging computer vision and mobile light detection and ranging (LiDAR) technologies to assess traffic sign retroreflectivity conditions. The proposed methodology includes a) an image-LiDAR registration method employing camera calibration and point co-planarity to register the 3D LiDAR point cloud with 2D video log images, b) a theoretical-empirical normalization scheme to adjust the magnitude of the LiDAR retro-intensity values with respect to LiDAR beam distance and incidence angle based on the radiometric responses, and c) a population-based retroreflectivity condition assessment method to evaluate the adequacy of a traffic sign retroreflectivity condition based on the correlation between the normalized LiDAR retro-intensity and the retroreflectivity values. For the proposed traffic sign detection methodology, comprehensive tests using representative datasets (e.g. with different road functions, data collection sources, and data qualities) were conducted to validate the performance of the two enhanced algorithms and the complete methodology. For the proposed retroreflectivity condition assessment methodology, the fundamental behavior of LiDAR retro-intensity was comprehensively tested and simulated under a controlled lab and roadway environment to quantify the impact of beam distance and incidence angle. A preliminary test on Type 1 engineer grade stop signs was conducted in the field to validate the performance of the proposed sign retroreflectivity condition assessment method. The results from both of the proposed methodologies are promising.