Generalized Image Recognition Algorithm for Sign Inventory
by Zhaozheng Hu, (Dalian Maritime University, Linghai Rd. No.1, Dalian 116026, P.R. China; and Georgia Institute of Technology, 210 Technology Circle, Savannah, GA 31407. E-mail: Zhaozheng.Hu@gatech.edu) and Yichang (James) Tsai, (Georgia Institute of Technology, 210 Technology Circle, Savannah, GA 31407. E-mail: James.Tsai@ce.gatech.edu)
Journal of Computing in Civil Engineering, Vol. 25, No. 2, March/April 2011, pp. 149-158, (doi: http://dx.doi.org/10.1061/(ASCE)CP.1943-5487.0000076)
Access full text
Purchase Subscription
Permissions for Reuse 
| Document type: |
Journal Paper |
| Abstract: |
Image detection and recognition algorithms are crucial for developing an intelligent sign inventory system using video log images. The technical challenge is to detect and recognize more than 670 different types of signs specified in the Manual on Uniform Traffic Control Devices (MUTCD). This paper develops a generalized image recognition algorithm that can differentiate various types of signs based on shape, color, location, probability distribution function (PDF), and Haar features trained and selected by the AdaBoost cascade method. Contributions of the paper are as follows: first, development of a generalized sign recognition algorithm instead of a sign-specific algorithm; second, development and incorporation of a new location PDF in the algorithm that describes the nonuniform distribution of actual sign locations in images; third, application and incorporation of the AdaBoost cascade method to automatically train and select Haar features; and fourth, validation of the proposed algorithm using real-world roadway video log images. The proposed algorithm has been tested with video log images collected on I-75 from Macon to Atlanta, covering 140 km of rural and urban roadways. The developed algorithm successfully recognized 28 out of 31 speed limit signs (a 90.3% recognition rate) and five false positives out of 136 images containing speed limit signs. These results show significant promise for development of an intelligent sign inventory system. With sufficient image training data sets, the proposed algorithm can be applied to other sign types. |
| Author Keywords: | 
| Sign inventory |
| MUTCD sign recognition |
| Generalized sign feature extraction |
| Video log images |
| Haar features |
|
