This study introduces a method for filtering lidar data based on a Support Vector Machines
(SVMs) classification method. Four study areas with different sensors and scene
characteristics were investigated. First, the Digital Surface Model (DSM) was generated for
the first and last pulses and then the differences between the first and last pulses (FP-LP) were
computed. A total of 25 uncorrelated feature attributes have been generated from the aerial
images, the lidar intensity image, DSM and FP-LP. The generated attributes were applied in
seven separate groups which include those from: Red, Green and Blue bands of the aerial
image; Intensity/IR image; DSM; FP-LP and the Total group of attributes. Finally, SVMs
were used to automatically classify buildings, trees, roads and ground from aerial images,
lidar data and the generated attributes, with the most accurate average classifications of 95%
being achieved. The Gaussian Radius Basis Function (RBF) kernel model was applied to find
the separating hyperplane for the SVMs classification.
A binary image was then generated by converting the digital numbers of roads and grass to
one while the digital numbers of buildings and trees were converted to zeros and all DSM’s
pixels which correspond to a pixel value of one in the binary image were interpolated into a
grid DTM. The interpolated DTM was then smoothed by a low-pass filter to remove low
vegetation and other objects which might be classified as ground.
After that the original 3D lidar point clouds was compared against the smoothed DTM and
labeled as ground or non-ground based on a predefined threshold of 30 cm.
To meet the objectives, the filtered data was compared against reference data that was
generated manually and both omission and commission errors were calculated. Further, we
evaluated the contributions of each group of attributes to the quality of the filtering process.
The results showed that the accuracy of the results was improved by fusing lidar data with
multispectral images regardless of the complexity of the terrain being filtered. |
