Cost-Effective Approach towards Building a Traffic Sign Data Inventory Using Open Street Images
-
2023-09-01
-
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final report, 3/1/22-9/30/23
-
Corporate Publisher:
-
Abstract:A computer vision algorithm has been applied to facilitate traffic sign detection and recognition tasks and solve labor-intensive issues. As public traffic signs datasets were collected in different regions, the existing models cannot be directly implemented into the US environment. Therefore, this study collected an additional 5,000 traffic signs in Washington data from Google Map API and self-installed dash cameras. In collaboration with Connected Cities with Smart Transportation (C2Smart) and the Washington State Department of Transportation (WSDOT), STARLab has collected the traffic sign data using three test vehicles equipped with onboard devices. The entire traveling route of three test vehicles covers most of the main roads in the Seattle region. Then, these data were manually labeled into 43 classes for training. To develop a traffic sign detection and recognition model (TSDR), the Faster R-CNN Inception V2 is selected as a base model for detection with an accuracy rate of 98.34%. Existing datasets and collected data were used to develop a customized traffic sign recognition model, which yielded an accuracy of 97.1%. In addition, an automated pipeline for traffic signs captures, detects, classifies, and stores is developed. By embedding the TSDR model in edge devices, this system allows the ability to uphold privacy standards. Processed data, consisting of an image and type of traffic sign, is transmitted to the server automatically. A sample traffic signs data inventory in Washington state was created. This database is a valuable resource for developing and testing various machine-learning models. Additionally, it can be used for asset management.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type: