Very unique paper about the feature detection with help of intensity information:
ADAPTATION OF ROADS TO ALS DATA BY MEANS OF NETWORK SNAKES
J. Goepfert, F. Rottensteiner
"ABSTRACT:
In the Authoritative Topographic Cartographic Information System (ATKIS®), which is the main public topographic data base inGermany, the heights and the 2D positions of objects such as roads are stored separately in the digital terrain model (DTM) anddigital landscape model (DLM). However, for many applications a combined visualization and processing of these two data sets is useful, leading to the demand for a 3D representation of the objects. For this purpose an integration of the height and position is essential. However, discrepancies exist between the DLM and the DTM due to different kinds of acquisition, processing, and modeling. This inconsistency has to be solved within the integration algorithm. Airborne Laserscanning (ALS) is used to acquire the height information for the DTM. Therefore, intensity values of the ALS echoes, which contain the reflectance properties of the lluminated objects, as well as object heights can be exploited in addition to the terrain height. However, the ALS data contain the information related to the DLM-objects only implicitly. Usually, features, for example edges or distinctive points, have to be extracted which can be connected to DLM-objects to realize the integration process. In this paper the matching of the data sets is realized by network snakes which are able to use the implicit ALS information about the position of the DLM-objects without the feature extraction step. The initialization of the contour with the vector data enables the use of ALS data as image energy for an iterative optimization process. Examples are given which apply the concept of network-snakes for the adaptation of a road network to ALS data taking advantage of the prior known topology.
CONCLUSION:
In this paper an active contour approach is applied to the
adaptation of a road network to ALS data. Even with a simple
definition of the image energy by combining ALS height and
intensity information promising results could be achieved.
Based on the exploitation of the topology of given initialisation
the snake converges to the desired position in comparison to the
ground truth. Contour parts connected by nodal points support
each other during the optimisation process.
The introduced algorithm offers many possibilities regarding
the integration of object knowledge for future research. For
example, constraints about the shape of the objects (parallelism
of road edges, slope constraints) or relations to other ATKIS
objects can be incorporated in the initialisation as well as in
additional energy terms. Furthermore, a more sophisticated
formulation of suitable image energies concerning different
vector objects can improve the applicability of the method.
Additionally, the algorithm can be extended to other DLM
objects having a relation to features in ALS data such as rivers.
Subsequently, all the adapted objects provide a dense network
of shift vectors which can be used in addition to prior accuracy
knowledge in order to improve the consistency of the DLM and
ALS data."
In the Authoritative Topographic Cartographic Information System (ATKIS®), which is the main public topographic data base inGermany, the heights and the 2D positions of objects such as roads are stored separately in the digital terrain model (DTM) anddigital landscape model (DLM). However, for many applications a combined visualization and processing of these two data sets is useful, leading to the demand for a 3D representation of the objects. For this purpose an integration of the height and position is essential. However, discrepancies exist between the DLM and the DTM due to different kinds of acquisition, processing, and modeling. This inconsistency has to be solved within the integration algorithm. Airborne Laserscanning (ALS) is used to acquire the height information for the DTM. Therefore, intensity values of the ALS echoes, which contain the reflectance properties of the lluminated objects, as well as object heights can be exploited in addition to the terrain height. However, the ALS data contain the information related to the DLM-objects only implicitly. Usually, features, for example edges or distinctive points, have to be extracted which can be connected to DLM-objects to realize the integration process. In this paper the matching of the data sets is realized by network snakes which are able to use the implicit ALS information about the position of the DLM-objects without the feature extraction step. The initialization of the contour with the vector data enables the use of ALS data as image energy for an iterative optimization process. Examples are given which apply the concept of network-snakes for the adaptation of a road network to ALS data taking advantage of the prior known topology.
CONCLUSION:
In this paper an active contour approach is applied to the
adaptation of a road network to ALS data. Even with a simple
definition of the image energy by combining ALS height and
intensity information promising results could be achieved.
Based on the exploitation of the topology of given initialisation
the snake converges to the desired position in comparison to the
ground truth. Contour parts connected by nodal points support
each other during the optimisation process.
The introduced algorithm offers many possibilities regarding
the integration of object knowledge for future research. For
example, constraints about the shape of the objects (parallelism
of road edges, slope constraints) or relations to other ATKIS
objects can be incorporated in the initialisation as well as in
additional energy terms. Furthermore, a more sophisticated
formulation of suitable image energies concerning different
vector objects can improve the applicability of the method.
Additionally, the algorithm can be extended to other DLM
objects having a relation to features in ALS data such as rivers.
Subsequently, all the adapted objects provide a dense network
of shift vectors which can be used in addition to prior accuracy
knowledge in order to improve the consistency of the DLM and
ALS data."
Image energy from a combination of intensity and height values (green: ground truth; blue: initialisation; red: final position)
No comments:
Post a Comment