Traffic signs are important transportation assets that are integrally related to roadway safety. Current traffic sign inventory by transportation agencies is still a manual process that is labor intensive and time consuming. A generalized traffic sign detection algorithm with a reasonably good performance has been developed using multiple features to automatically detect signs. However, correctly detecting signs with discontinuous boundaries remains a challenge that lead to false negatives. To address the issue,a partial differential equation (PDE) method, which is a region-based active contour model, is proposed. A new energy function incorporated with a location probability distribution function (PDF) is proposed for the active contour. The contributions of this paper include the following: 1) proposes a PDE method to improve discontinuous sign detection; 2) proposes a location PDF incorporated energy function component to provide a more accurate contour initial guess; 3) the new method is integrated into existing system framework without affecting existing methods. The video-log image data from the City of Nashville are used to perform the test. Using the new method, 24 out of 26 images with a discontinuity issue are correctly detected, and 607 images from a road segment are tested to compare the performance of the enhanced system and the previous system. Results show that falsenegative rate has been reduced 6.8% from 21.1% while the false positive rate has increased 1.2% from 11.2% with the enhanced system. The results show that incorporating the proposed PDE method to develop an enhanced sign detection system is promising. Future research for incorporating additional distinct features into the generalized algorithm is also discussed.
Samenvatting