We are involved in a research of composing a self-positioning system of AGV(Automated Guided Vehicle). This system has two components, that are saliency detection part and machine learning part. We report the process of the latter of the two. We use SVM(Support Vector Machine) to execute the machine learning of landmark images. To use SVM, We extracted SURF(Speed-Up Robust Features) features from landmark images, and grouped them depending their similarity. The centroid of each cluster is treated as VW (Visual Words). Thus, these images are transformed into VW histograms. By using prior information of landmarks, that have been accumulated during the process of machine learning, the saliency detection regions are focused on the part where learned landmarks are estimated to exist in the environmental images. In this report, We are going to discuss the application of soft margin SVM(Support vector machine) into landmark detection.