Search for: All records

Background: The rise in work zone crashes due to distracted and aggressive driving calls for improved safety measures. While Truck-Mounted Attenuators (TMAs) have helped reduce crash severity, the increasing number of crashes involving TMAs shows the need for improved warning systems. Methods: This study proposes an AI-enabled vision system to automatically alert drivers on collision courses with TMAs, addressing the limitations of manual alert systems. The system uses multi-task learning (MTL) to detect and classify vehicles, estimate distance zones (danger, warning, and safe), and perform lane and road segmentation. MTL improves efficiency and accuracy, making it ideal for devices with limited resources. Using a Generalized Efficient Layer Aggregation Network (GELAN) backbone, the system enhances stability and performance. Additionally, an alert module triggers alarms based on speed, acceleration, and time to collision. Results: The model achieves a recall of 90.5%, an mAP of 0.792 for vehicle detection, an mIOU of 0.948 for road segmentation, an accuracy of 81.5% for lane segmentation, and 83.8% accuracy for distance classification. Conclusions: The results show the system accurately detects vehicles, classifies distances, and provides real-time alerts, reducing TMA collision risks and enhancing work zone safety. 

The analysis of factors that influence the occurrence of roadway crashes within a specified locality have his- torically been reliant on the assessment of physical infrastructure, historical crash frequency, environmental factors and driver characteristics. The consensus over the years has been drawn to the idea that human factors, specifically regarding driving behaviors, account for the majority of crash outcomes on our roadways. With the emergence of connected vehicle data in the last few years, the capacity to analyze real time driving behavior has become a possibility for safety analysts. Driving volatility has emerged as a valuable proxy for driving behavior and indicator of safety. In this study, evidence of the spatial relationship between driving volatility and historical crash hotspots is uncovered. Utilizing an entropy-based analysis, this study discovered generally strong positive spatial relationships between locations of volatile driving events and historical crashes, with R2 values ranging from 0.015 to 0.970 and a mean of 0.612 for hard accelerations, and 0.048 to 0.996 and a mean of 0.678 for hard decelerations. Including temporal context presented insights showing that the relationships are significant for over 60 % of the coverage area usually between the hours of 7 am to 7 pm, with average R2 values of 0.594 for hard accelerations, and 0.629 for hard decelerations. 

Object recognition and depth perception are two tightly coupled tasks that are indispensable for situational awareness. Most autonomous systems are able to perform these tasks by processing and integrating data streaming from a variety of sensors. The multiple hardware and sophisticated software architectures required to operate these systems makes them expensive to scale and operate. This paper implements a fast, monocular vision system that can be used for simultaneous object recognition and depth perception. We borrow from the architecture of a start-of-the-art object recognition system, YOLOv3, and extend its architecture by incorporating distances and modifying its loss functions and prediction vectors to enable it to multitask on both tasks. The vision system is trained on a large database acquired through the coupling of LiDAR measurements with complementary 360-degree camera to generate a high-fidelity labeled dataset. The performance of the multipurpose network is evaluated on a test dataset consisting of a total of 7,634 objects collected on a different road network. When compared with ground truth LiDAR data, the proposed network achieves a mean absolute percentage error rate of 11% on the passenger car within 10 m and a mean error rate of 7% or 9% on the truck within 10 m and beyond 10 m, respectively. It was also observed that adding a second task (depth perception) to the modeling network improved the accuracy of object detection by about 3%. The proposed multipurpose model can be used for the development of automated alert systems, traffic monitoring, and safety monitoring. 

This study proposes a data fusion and deep learning (DL) framework that learns high-level traffic features from network-level images to predict large-scale, multi-route, speed and volume of connected vehicles (CVs). We present a scalable and parallel method of processing statewide CVs’ trajectory data that leads to real-time insights on the micro-scale in time and space (two-dimensional (2D) arrays) on graphics processing unit (GPUs) using the Nvidia rapids framework and dask parallel cluster, which provided a 50× speed-up in the data extraction, transform and load (ETL). A UNet model is then applied to perform feature extraction and multi-route speed and volume channels over a multi-step prediction horizon. The accuracy and robustness of the proposed model are evaluated by taking different road types, times of day and image snippets and comparing the model to benchmarks: Convolutional Long–Short-Term Memory (ConvLSTM) and a historical average (HA). The results show that the proposed model outperforms benchmarks with an average improvement of 15% over ConvLSTM and 65% over the HA. Comparing the image snippets from each prediction model to the actual image shows that image textures were highly similar in UNet to the benchmark models used. UNet’s dominance in performing image predictions was also evident in multi-step forecasting, where the increase in errors was relatively minimal over longer prediction horizons.