This paper presents an approach to detecting laterals which consists of three steps. First, pipe images are restored and enhanced by implementing image processing techniques. Second, gray-scale morphology, anisotropic diffusion filters and histogram thresholding are performed to segment candidate laterals. In the third phase, AdaBoost is used to classify candidate laterals and its performance is compared to Support Vector Machine and K-NN. Experimental results show that AdaBoost with twenty weak classifiers outperform other algorithms. Our approach achieve about 90% test accuracy and has been tested on pipelines of 10000 meters in length or about 6000 scanned images of real sewer pipes from various cities all over the world.