United States Department of Transportation
i
Intersection Safety for the Vulnerable
-
2024-09-19
-
[PDF-718.09 KB]
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final Report (July 1, 2023-June 30, 2024)
-
Corporate Publisher:
-
Abstract:The goal of the proposed work is the development of computational methods that can be used towards enhancing the safety of vulnerable road users (VRUs) at intersections. To accomplish this goal, we envision a cyber-physical system that detects VRUs, calculates a vulnerability score (based on spatial and temporal risk), then takes appropriate action or actions to minimize the opportunity for injury. The core of the system is based on automatically detecting VRUs in visual data captured from cameras. Accomplishing this task for automation requires a couple of core components. First, understanding the scene to assess vulnerability requires real-world measurements. Unfortunately, outdoor security and traffic cameras are challenging to keep calibrated, and therefore, they do not have any calibration data required for traditional 3D computational methods. To address this issue and take advantage of the widespread use of outdoor cameras, we have developed a method to automatically calibrate any outdoor camera from street-level images. Second, detecting people with vulnerabilities requires an annotated dataset of VRUs, e.g., person walking with cane, user of a wheelchair, bicyclist, etc. There are a handful of datasets publicly available, but not nearly enough with annotated VRUs. To fill the dataset gap, we have developed a framework to use time-lapse videos from stationary cameras to synthesize realistic scenarios (with and without occlusion) by extracting unconcluded objects and compositing them back into the background image at their original positions. Using this method, public datasets can be augmented with the generated photorealistic synthetic images.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
