Vehicle classification based on line fitting to a vehicle profile with DP-matching

摘要

We propose a vehicle classification method based on shape models and line fitting, which is for a vehicle profile obtained with an optical rangefinder over a road. To be robust against changes of shape; each model of vehicle class is expressed with several line segments, and degrees of line fitting to partial profile corresponding to the segment are evaluated on principal component analysis. In this process, DP-matching is applied to get whole evaluation values and select the class having the best score. Various shapes of vehicle can be classified efficiently, as the models permit fixed deformation. An experiment result with real data shows that the method is applicable to practical use.