Introduction to my Master's Thesis

Topic: Classification in Urban Areas based on LiDAR and Aerial Colour Infrared Imagery

LiDAR data has become more accessible during last few years. Almost every big city is covered yearly or every few years with both aerial photos and LiDAR data. When the case is digital camera both visible and near-infrared wavelengths are recorded. Photographs are used to generate orthophotomap, which can be used parallel to traditional map. LiDAR data usually helps to build Digital Surface Model (DSM) and as a result of filtration Digital Elevation Model (DEM). Sometimes mentioned data may be used to generate 3D model of the city.

This study shows another application of the data. The result will be classification of urban area extracting objects such as: buildings, different types of vegetation, roads etc. It is highly difficult to reach high accuracy object classification based only on spectral information of objects. This is because urban area is spectrally varied: the same type of object may have different spectral response. Below there is an example based on buildings:

Spectral information about buildings is inconsistent. Reasons:

- Roofs are made from different materials, from coloursed roof tiles, to metal roofs.

- There are different types of roofs: flat, gable etc. There are shadows, which change the spectral characteristic of parts of the roof.

- Spectral information is different on different photographs (dependent on exposition settings, initial processing etc.)

Example for different types of buildings (source: Definiens AG: Guided Tour Simple Example of Building Extraction using LiDAR)

In contrast to spectral information, elevation information is stabile for buildings. Buildings are elevated to their surrounding. At the edges of a building there is a very steep slope. The elevation changes suddenly.

The DSM image layer (source: Definiens AG: Guided Tour Simple Example of Building Extraction using LiDAR)

In conclusion the usage of both spectral and elevation information may improve the results of classification to the level that wasn’t possible before. It will be possible to distinct even single buildings and trees with high spatial and thematic accuracy.