United States Department of Transportation
i
Crash Prediction and Avoidance by Identifying and Evaluating Risk Factors from Onboard Cameras
-
2020-09-01
-
[PDF-3.84 MB]
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final Report January 2019-July 2020
-
Corporate Publisher:
-
Abstract:Motor vehicle crashes are a huge concern of roadway transportation safety, resulting in over 37,000 fatalities and $800 million losses annually. In recent years, the number of road fatalities has been growing. Traditionally used and identifiable risk factor explanations no longer fully account for the causes of a recent increase in road fatalities. Human beings have bounded abilities in vision, cognition, making judgment, and simultaneously handling multiple tasks, particularly in complex, dynamic environments or in response to sudden situations. Therefore, assisting them in cognition of risks and making the right decisions rapidly in a near real-time manner is in need to advance transportation toward a zero fatality rate. This project’s motivation is to develop a data-driven, computer-vision (CV) empowered, verifiable system that can predict crashes, and thus improve drivers’ ability to avoid them. Pursuing a systematic approach, this project seamlessly integrates data analytics, deep learning, computer vision technology, and a rigorous verification process to achieve the goal. Specifically, this project creates a spatio-temporal attention guidance for CV-based crash risk assessment through analyzing fatal crash report data retrieved from Fatality Analysis Reporting System (FARS). The guidance informs the likelihood of crash and crash types given the time and location information of driving scenes, thus giving the driving scene analysis a clear focus. Then, a system of deep neural networks is developed to perform a driving scene analysis in support of crash risk assessment and prevention. The scene classification result allows for retrieving the relevant guidance for crash risk assessment and prevention. The joint results from the object detection and drivable area segmentation help identify risky pedestrians and vehicles in the surrounding traffic. Evaluation and examples demonstrate the effectiveness of the proposed technologies.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
