Context based multiple railway object recognition from mobile laser scanning data

In this paper, we present a context based multiple railway object recognition method from mobile laser scanning data. This research makes use of contextual information for classification, which is retrieved from the unlabeled neighborhood as feature vector. The interaction (object context) among object labels is also utilized to enforce local smoothness constraint. The model we use to incorporate contextual information is Conditional Random Field (CRF). By maximizing the object label agreement in the local neighborhood, CRF could improve the classification results obtained from local GMM-EM classifier. The proposed method was validated with mobile laser scanning data using cross validation.