LiDAR Perception and Evaluation Method for Road Traffic Marking Retroreflection

Abstract

To solve the problem of low efficiency in retroreflection maintenance of road traffic markings, in this study a vehicle-mounted LiDAR-based perception and evaluation method of retroreflection of markings is proposed. First, this method establishes a calibration prediction model for LiDAR based on a regression decision tree. Then, a marking maintenance evaluation model is constructed in combination with the decision threshold proposed by China’s national standard, and the accuracy of the maintenance evaluation model is analyzed using F1-score, recall, and precision. In this study, four marking lines were used as the calibration data source, and a dataset of an independent 1,300 m road section was used to verify the established models. The results show that the coefficient of the retroreflected luminance ( RL) and the reflection intensity of the markings are positively correlated. During the construction of the calibration prediction model, the multiple linear regression functions, the second-order polynomial functions, and the decision tree are compared, and the result indicates that decision tree has the best fit to the data with the coefficient of determination for the established calibration prediction model better than 0.95. The agreement between the maintenance decision obtained from the maintenance evaluation model and the traditional method is more than 85%. The time cost is reduced by at least 90%. The proposed calibration prediction model can accurately predict the RL, and can quickly collect the RL values of the in-service road traffic markings. The proposed maintenance evaluation model is highly efficient and can replace the traditional evaluation method for markings.

Traffic Marking Extraction Algorithm Based on Image and Point Cloud Data

“…Due to the different calculation methods of echo intensity by different LiDAR devices, there are also differences in the results. Therefore, this study applies the maximum inter class variance method to adaptive threshold segmentation processing of point cloud data [25][26]. The Otsu algorithm mainly divides pixels into two parts: background pixels and target pixels by setting pixel grayscale threshold values.…”

Improving Autonomous Vehicle Perception through Evaluating LiDAR Capabilities and Handheld Retroreflectivity Assessments

“…The literature has extensively discussed the effectiveness of LiDAR in detecting roads, traffic signs, and markings in AV environments [ 13 ]. Intensity analysis, a crucial method for optimizing traffic sign visibility through LiDAR, is essential for determining visibility and contrasting various environmental conditions [ 14 ]. Moreover, paper [ 15 ] referenced the work of [ 16 ], which introduced a LiDAR point cloud reflectivity method for detecting and categorizing vertical traffic signs.…”