This project developed and applied pattern recognition and image processing algorithms to automatically detect traffic signs in roadway video log images, thereby reducing the number of video log images needing to be manually reviewed. The algorithm development was divided into two parts: sign detection and sign recognition. A robust algorithm based on multi-feature fusion was proposed for detecting signs.The algorithm performed the steps of training and testing. In the training step, characteristics of MUTCD signs (including shape, color distribution, location distribution, width-height ratios, and others) in video log images were analyzed. For each feature, one or more sign detectors were designed, and their parameters (such as threshold values) were adjusted. Next, efforts were directed at developing a sign recognition algorithm capable of classifying a variety of sign images using the Adaboost Cascade algorithm. This algorithm also consisted of two main steps: training and testing. The algorithm was tested with video log images collected on I-75 from Macon to Atlanta, Georgia, covering 140 km of both rural and urban roadways. The algorithm successfully recognized 28 of 31 speed limit signs (a 90.3% recognition rate) and had only 5 false positives out of 136 speed limit sign images. With sufficient image training data sets, the proposed algorithm should also be applicable to other types of signs. The proposed algorithms for both detection and recognition appear very promising for developing an intelligent sign inventory system.