A Novel Real-Time Autonomous Crack Inspection System Based on Unmanned Aerial Vehicles.

Traditional methods on crack inspection for large infrastructures require a number of structural health inspection devices and instruments. They usually use the signal changes caused by physical deformations from cracks to detect the cracks, which is time-consuming and cost-ineffective. In this work, we propose a novel real-time crack inspection system based on unmanned aerial vehicles for real-world applications. The proposed system successfully detects and classifies various types of cracks. It can accurately find the crack positions in the world coordinate system. Our detector is based on an improved YOLOv4 with an attention module, which produces 90.02% mean average precision (mAP) and outperforms the YOLOv4-original by 5.23% in terms of mAP. The proposed system is low-cost and lightweight. Moreover, it is not restricted by navigation trajectories. The experimental results demonstrate the robustness and effectiveness of our system in real-world crack inspection tasks.

Determinants and Prediction of Injury Severities in Multi-Vehicle-Involved Crashes.

Multi-vehicle (MV) crashes, which can lead to great damages to society, have always been a serious issue for traffic safety. A further understanding of crash severity can help transportation engineers identify the critical reasons and find effective countermeasures to improve transportation safety. However, studies involving methods of machine learning to predict the possibility of injury-severity of MV crashes are rarely seen. Besides that, previous studies have rarely taken temporal stability into consideration in MV crashes. To bridge these knowledge gaps, two kinds of models: random parameters logit model (RPL), with heterogeneities in the means and variances, and Random Forest (RF) were employed in this research to identify the critical contributing factors and to predict the possibility of MV injury-severity. Three-year (2016-2018) MV data from Washington, United States, extracted from the Highway Safety Information System (HSIS), were applied for crash injury-severity analysis. In addition, a series of likelihood ratio tests were conducted for temporal stability between different years. Four indicators were employed to measure the prediction performance of the selected models, and four categories of crash-related characteristics were specifically investigated based on the RPL model. The results showed that the machine learning-based models performed better than the statistical models did when taking the overall accuracy as an evaluation indicator. However, the statistical models had a better prediction performance than the machine learning models had considering crash costs. Temporal instabilities were present between 2016 and 2017 MV data. The effect of significant factors was elaborated based on the RPL model with heterogeneities in the means and variances.

Vehicle Detection under Adverse Weather from Roadside LiDAR Data.

Roadside light detection and ranging (LiDAR) is an emerging traffic data collection device and has recently been deployed in different transportation areas. The current data processing algorithms for roadside LiDAR are usually developed assuming normal weather conditions. Adverse weather conditions, such as windy and snowy conditions, could be challenges for data processing. This paper examines the performance of the state-of-the-art data processing algorithms developed for roadside LiDAR under adverse weather and then composed an improved background filtering and object clustering method in order to process the roadside LiDAR data, which was proven to perform better under windy and snowy weather. The testing results showed that the accuracy of the background filtering and point clustering was greatly improved compared to the state-of-the-art methods. With this new approach, vehicles can be identified with relatively high accuracy under windy and snowy weather.