Today, the analysis of 3D point clouds acquired with LiDAR or photogrammetry technique has become an operational task for mapping as well as for infrastructure and environmental monitoring.
Numerous applications require the identification of the point cloud and their properties. So far, most software have been focusing on the analysis of constructed objects versus natural landscapes.
Automatic classification allow to categorize points into different classes like roads, buildings, high/medium/low vegetation, railroads, wires, water, misc. Man-made objects, etc. Without this capability, hours of tedious work would be necessary to edit the point cloud and manually identify the points with accuracy.
But despite the evolution of available software using machine-learning, the automatic extraction of accurate information from LiDAR point clouds still remains a challenge.
Among the processing methods, classifying the LAS data into categorical object instances is the first and most critical step for further data processing. These existing classification strategies can be categorized into three groups: point-based classification, segment-based classification and multiple-entity-based classification.
In this process, features of individual points are firstly extracted. Then, a classifier is trained, using a number of selected training samples. Finally, the input point cloud is classified via the trained classifier and the extracted features.
Additionally, to compute the features of individual points, a respective neighborhood definition is required to describe the local 3D structure around each individual point.
The scale parameter, either a fixed radius or a constant value, is required. Due to the variation of local 3D structures and point densities, the constant value often fails to describe the local structural configurations. Thus, more and more studies focus on seeking an optimal neighborhood size for each individual point. Unfortunately, these neighborhood optimization methods require repetitive calculations for each point, therefore they are rather time-consuming, which is the main disadvantage of this kind of classification.
Generally, segment-based classification methods first perform segmentation on the point
cloud after removing the ground points. Then, the non-ground points are divided into a
number of segments, and features are extracted for each of them. Finally, a fuzzy model classifier or several classification rules are used to classify the segments.
Segment-based classification relies heavily on its employed segmentation method.
Various segmentation methods allow to segment input 3D point clouds into only one type of geometric structure. Actually, point clouds consist of a variety of geometric structures, such as planes, smooth surfaces and rough surfaces. In a complex 3D scene, there may exist regular and irregular man-made objects, and natural objects. Regular man-made objects such as buildings are composed of planar surfaces and smooth surfaces, while irregular man-made objects such as cars and natural objects like trees are composed of rough surfaces.
Despite those limitations, segment-based methods still have two main benefits in contrast to point-based classification:
- segments are helpful to compute geometric features which facilitate the neighborhood optimization
- segments give several new attributes which are helpful to employ semantic rules.
Multiple-entity-based classification is considered as a combination of the segment-based and point-based classification. To solve the problem that a complex 3D scene is difficult to be characterized by only individual points or one kind of segments, this method utilizes three kinds of entities: points, planar segments, mean shift segments. In the process of classification, the input LAS point cloud is first divided into ground points and non-ground points. Next, planar segments are extracted from the non-ground points, and the scattered points are remained. Then, the planar segments are classified into several classes. The remained points are point-wise classified based on the contextual information offered by the classified planar segments. Finally, in complex areas where vegetation covers building roofs, mean shift segments are extracted to classify these areas.
However, the process of this method is an hierarchical classification procedure, which involves many steps. Besides, the mean shift segments and planar segments are derived from different segmentation methods, which adds additional classification steps. To simplify the classification process, a point cloud segmentation method that is able to extract more than one kind of segments is required.
It can be concluded that segment-based classification of point clouds is a promising alternative to point-based classification, keeping in mind that the analysis of natural objects is especially challenging, since their identification is sometimes ambiguous and their boundaries are often imprecise and may overlap one on another.
Because airborne LiDAR systems use remote sensing technologies -including PPK/RTK GNSS and IMU data-, it allow them to do more analysis of the returned laser energy, and record the useful following attributes:
- Height above ground level and intensity
- Differences of elevations
- Standard deviations of elevations
- Standard deviations of intensity values
- Angle of planar surface normal
- Principal component percentage of variance
- Point density
Terrestrial laser scanners use different sorts of sensors that do not provide the additional information needed to easily analyze and classify point clouds.
While the LAS point cloud format, traditionally used by airborne LiDAR systems, is capable of supporting classified point clouds, simply saving terrestrial LiDAR point clouds into the LAS format does not deliver the classified point clouds that users generally need.