Building footprint extraction using LiDAR data and other classification techniques for a segment of atlantic county, New Jersey
Student name: Ms R. Prerna
Guide: Mr Chander Kumar Singh
Year of completion: 2012
Host Organisation: Risk Management Solutions India Pvt. Ltd.
Supervisor (Host Organisation): Ms Bina Shetty
Abstract: In a study of any urban landscape, it is of utmost importance to have an accurate
inventory of all buildings, with their attributes like shape, area, geometry, building
material, roof age etc in order to carry out planning and development. Over the past
few years, the need for acquiring high quality building footprints for major parts of
the world has been increasing. This is primarily due to the role they play in multiple
sectors of planning such as town/city planning, urban growth assessment,
telecommunication, transport network planning, impervious surface area
calculations, risk/disaster management, insurance coverage and many more.
Here at Risk Management Solution India Pvt. Ltd., a major part of the
company’s work plan is focused on the accurate assessment of building information
across United States in order to deliver this information to clients of Insurance
companies and help them insure the lives of millions of people. There are various
ways in which this information is captured, most of which are extremely timeconsuming
and tedious to perform.
In the current study, the main objective has been to acquire building
footprints using aerial LiDAR data sets with the use of a LiDAR processing software
named LP360 (Advanced version) which enables the user to extract building
footprints automatically. The methodology involved filtering of the data sets to
attain a meaningful classification followed by feature extraction in the form of ESRI
shapefiles along with associated attributes.
The parameters fed into the software have been able to produce good
outputs as they seem to fit quite proportionately with aerial photographs of the study
area. These parameters with minor modifications can be successfully employed for
footprint capturing in other areas as well due to the robustness of the software.
Due to the different acquisition dates of the aerial photographs (2007) and
the LiDAR datasets (2010), the building footprints acquired show buildings in
regions which appear to be open land in the aerial photographs. These additional
buildings have been removed from the outputs in order to maintain a degree of
consistency between the two data sets and also for comparison purposes.
Also, high resolution aerial photographs were acquired for the same area
which, were classified employing the concept of Object Oriented Classification
(OOC). Herein the buildings were first clubbed together as “objects†meaning a
group of pixels having homogenous colour/tone and shape and similar objects were
extracted from the photographs resulting in building footprint segments. This
methodology of acquiring building footprints has also been successfully applied
giving precise outputs.
Apart from the above two mentioned methods, the third technique for
footprint extraction was to derive footprints from Digital Surface Models generated
from LiDAR point clouds by means of interpolation and subsequently vectorizing the
building polygons. This has also been a valid approach towards feature extraction
but the accuracy of this method has been the least which shall be discussed further
in depth, nevertheless it gave reasonably good outputs.
Valid comparisons of the footprints collected by the aforementioned
techniques with ground data has helped in the true evaluation of the procedures
followed making them valuable techniques of footprint extraction.
Keywords:
Light Detection and Ranging (LiDAR); Building footprints; Object Oriented
Classification (OOC); Image Segmentation; Digital Surface Model (DSM)
