In this paper we describe a mobile-based hazardous material (hazmat) sign detection and recognition system. Hazmat sign detection is based on visual saliency models. We use saliency maps to denote regions that are likely to contain hazmat signs in complex scenes and then use a convex quadrilateral shape detector to find hazmat sign candidates in these regions. Experimental results show that our proposed hazmat sign detection and recognition method is capable of dealing with projective distorted, blurred, and shaded signs. The test image dataset consists of images taken in the field under various lighting and weather conditions, distances, and perspectives.