Automatic Segmentation and Shape-Based Classification of Retro-Reflective Traffic Signs from Mobile LiDAR Data

Abstract

Recently, many studies have demonstrated the valid contribution of mobile laser scanning to road safety improvements, thus intense efforts have been made to implement automatic data processing using laser scanning data, with special emphasis on road object recognition. This study is focused on the detection and classification of retro-reflective vertical traffic signs according to their function (danger, give way, prohibition/obligation, and indication) from mobile laser scanning data by considering geometric and radiometric information, but without relying on trajectory data. The global strategy for segmentation involves the application of an optimized intensity threshold in order to segment the points that correspond to traffic sign panels. Next, contour recognition is performed for each sign using a linear regression model based on a raster image, which is generated for each cluster of points. The shape evaluation is motivated by the correspondence between contour shape and function of the traffic. The completeness of results for detection (92.11%) and classification (83.91%) demonstrates that this implementation is promising for the automatic detection and inventory analysis of traffic signs in road mapping applications. The efficiency rates are acceptable in urban areas, but our tests indicate that the detection and classification rates are more robust in road environments.