Vehicle recognition from multi-spectral and LiDAR elevation data with object-oriented analysis
This research study addresses issues of transport modeling aiming to the recognition of vehicles, with perspective to generate a knowledge base for an automatic recognition of vehicles from similar type of data. Based on two datasets consisted of multi-spectral data of an RGB/NIR line scanner, intensity data and elevation data of a LiDAR scanner, the goal was achieved with an object-oriented and fuzzy logic approach. Nevertheless, an object-oriented image analysis comprises of multiple complex dependencies that is required to be reduced for the achievement of an optimum accuracy in classification. Namely, via a series of tests was attempted to build simplified hierarchies and class descriptions that additionally would act in compliance with the criterion of homogeneity. Regarding this criterion and due to the shape and diminutive size of the object in question, a classification would be rendered inaccurate without the utilization of the most possible colour criterion and the most necessary shape criterion. The process of object-oriented analysis included three levels of segmentation that were later classified based on descriptions drawn up for the photointerpretation classes, which were recognized in the two datasets that were provided. Initially, the data were processed and analyzed with two approaches based on object-oriented analysis, and the resulted classes were described with fuzzy logic. From the “top-down” approach large surfaces of various land uses and several types of vehicles were ensued, avoiding misclassification of lorries. From the second, “bottom-up” approach, various types of vehicles emerged as compact objects, after merging the resulted objects that were precisely describing parts of the actual objects of classes. With focal point the vehicles extraction, the description of the vehicle class was assigned without significant overlap in the range of features used to describe other classes, rendering to a greater segment classification accuracy of the objects that were depicting vehicles. Despite efforts to produce suitable segments during the segmentation of this level, due to the size of the vehicles in the given data resolution and of the errors resulting from the general classification, vehicle shape could not be properly defined in extent. The developed methodology can be serve as a guide for recognizing vehicles both in urban and industrial environment from the same resolution of multi-spectral and LiDAR elevation data, by effectuating minor alterations in the range of features that has been employed.