RICERCA ITALIANA     

Research program

The geomatics in support of the actions of Government of the territory

University Co-ordinator

Università degli Studi di NAPOLI "Parthenope" - SCIENZE APPLICATE - ()

Research Unit Leader

Salvatore Troisi

Description

Until a few years ago, the three-dimensional representation of urban zone played a main role of visualisation, providing a better understanding of its geometry, whereas it was not used as a tool to investigate the territory comparable to the numerical cartography; it was seen mainly as a product dedicated to creating scenic effects respect to a tool of supporting technical investigations.
Nowadays the request of three-dimensional models comes from technicians who use them for different purposes such as the planning of mobile phone networks and similar technical plants, the study of the traffic investigations concerning safety aspects, and so on.
The three-dimensional description of the area and anthropical structures is becoming a fundamental tool for local authorities who are responsible for their management and control.
The “3D city model” is an excellent support for the 3D representation and consultation of all the information related to terrain and buildings allowing to display and manage as maps as both heights and volumes of urban zones .
In the 3D modelling pipeline the use of data from a specific survey could be not sufficient: aerial laser and photogrammetric data are able to provide most of the metric and semantic information concerning the surveyed objects, but they can not overcome the natural or artificial obstructions which are interposed between the sensor and the object to be detected.
Tools such as “Pictometry” were planned to provide different points of view compared to pseudonadiral ones of classical photogrammetry: “Pictometry” is able to partially overcome the obstructions and take a new step towards a detailed modelling and allowing a better control of teritory and buildings.
Similarly, a survey performed with terrestrial photogrammetry or terrestrial laser scanning allows to overcome the problems due to the obstructions which occlude the sight from an aerial view and they consent to reach a high definition as regards the external facades of a building (including those characterized by historical and complex architectures), providing, in this way, to highlight some particulars such as balconies, verandas, and the re-entry of the walls respect to the eaves. In the contrary it does not supply any information about the shape of roofs or any superstructure present on the terraces of the buildings themselves (including any abuse in the construction).
The cameras or digital videocameras can also operate from terrestrial vehicles which may be equipped , as for the aircrafts, with systems capable of providing the location and the attitude of the cameras.
These systems, known with the acronym MMS (Mobile Mapping Systems), have already been tested successfully in the field of cadastre road.
The aim of this research is to develop some innovative approaches for the processing of different kinds of data (laser data from air and land, GPS-INS, terrestrial and aerial imagery, topographic and cartographic data) for the generation of three-dimensional buildings’ models.
The phases of the program are as follows:

1. Development of algorithms for the automatic modelling of buildings’ roofs starting from Airborne Laser Scanning (ALS) data or aerial photogrammetry also by using additional information such as images and numerical cartography.
Till a few years ago, the generation of buildings’ models was possible only through manual measurements made by expert operators by using analytical and digital photogrammetric stations.
Nowadays the new coming technologies like the ALS system together with new software procedures, which can work on common PC desktop, have opened new perspectives in the 3D modelling. The research in the buildings extraction field from ALS data, has taken many step forward. Two main procedures are generally used: the first one works on single buildings, the second one on the pitches.
In the first case additional data are used (besides the point cloud provided by the ALS system), such as the ones extractable from numerical cartography to split data in as sub-regions as the number of buildings. These sub-regions which have to be processed, one by one, successively. In the second procedure pitches are automatically detected independently from the buildings they belong to; however such a procedure needs a large work of post-processing to regularize and join together roofs’ pitches, becoming a time consuming technique similar to the stereo-plotting one.
The aim of the research unit is the development of algorithms for modelling, in a semi-automatic way (without any use of graphic editing), the roofs of buildings from ALS data or high-density points clouds extracted with image matching techniques. The research group has already dealed with these problems in the past year related to automatic segmentation and identification of eaves lines with satisfactory results. Part of the work will be dedicated to improve the obtained results and to integrate them with terrestrial data (images, video frame, “Terrestrial Laser Scanning” TLS).


2.Conversion of the algorithms for filtering ALS data in order to use them in the case of surveys with TLS.
In the estimation of the buildings’ volumes it can be often necessary to have a digital terrain model (DTM) from which, up to the roofs level, to calculate the volume.
The generation of such models can be made by means of manual and automatic photogrammetric measurements or through the use of suitable procedures of filtering airborne laser scanning data.
The research unit has already developed, experimented and carried out successfully in the past years several software procedures for the processing of airborne laser scanning data; in this specific case, to divide point belonging to the ground from those surveyed on the overlooking anthropical objects. Such a procedure will be also adapted to the case of point clouds extracted by means of image matching in digital photogrammetry.
As above described, in some cases the three-dimensional data, due to possible obstructions, can be not sufficient to obtain complete information about buildings.
For this reason an integration with terrestrial surveying techniques is required.
The research unit wants to identify analogies and differences between the filtering of ALS and TLS data in order to use the existing own algorithms for filtering digital surface models of facades if terrestrial surveys were needed.

3. Development of calibration procedures to integrate GPS/INS system with digital cameras and video cameras;
The use of GPS/INS system is needed for the estimation of image exterior orientation parameters and for laser scanning systems when these instruments are positioned on moving vehicles both aerial and terrestrial.
In this last case the problems concerning with its employment in urban areas are very numerous addressed to the acquisition of satellite signal. Partial visibility of the sky, obstructed by the buildings, together with the multi-reflections (multipath) of the signal, generated by the neighbouring objects, are the main causes.
The use of an integrated GPS/INS system in the research project rises from the need of knowing approximate values of image exterior orientation parameters to be introduced as known values in suitable procedures of mathematic and image analysis.
These procedures include the automatic measurements of tie points (homologue points in two or more images), the scaling of relative photogrammetric models, the automatic recognition of homologue geometric entities on the images.
Such a system has to be calibrated appropriately, by means of sophisticated algorithms, owing to the fact that the system’s sensors have different instrumental centre positions and orientations which are not measurable topographically with the necessary approximation.
The research unit will analyse the problems concerning with such systems developing algorithms of auto-calibration.

4. Development of image-processing algorithms for the automatic recognition of homologue geometric entities in two or more frames in order to recover both in real-time and in post-processing the camera station positions.
The relative and the absolute image orientation, consequently the recovering of camera station positions, are the first, and thus very important step, within the 3D image-based modelling pipeline.
The relative image orientation can be seen as a problem of seeking a certain number of homologue points in two or more images. Till now such a type of operation has been made especially by means of manual and semi-automatic measurements with an automation degree which depends on the kind of photogrammetric application.
In the aerial case some algorithms for locating in automatic way homologue points have already been made (with the only condition that some distinguishable points are visible on the images, as it is in the case of urban areas). In the standard aerial case, images are taken along the flight line with an almost regular configuration of camera station positions and an overlapping area visible in two subsequent images. The algorithms seek for points which are well distinguishable such as the points with a high local contrast or further parameters. The algorithms for image points correspondence between two subsequent images take advantage of the a priori information coming from the flight plan.
In the terrestrial case, the procedures of automatic image orientation consist in recognizing coded targets previously located on the object to be surveyed. The image points correspondence occurs by means of the different coded targets.
The employment of “interest operator” procedures for the automatic points marking meets with many difficulties owing to the network design of camera stations and to the characteristics of the object to be surveyed. Nowadays the commercial software provide for automatic image points marking but only for aerial camera configurations; the use of these software in the terrestrial case, with tilted images, generally leads to a failure in the matching.
The research unit wants to develop algorithms for automatic image orientation intentionally developed for images taken with digital cameras and video cameras by using, if it was needed, information coming from additional sensors like GPS/INS.

5. Development of methods to georeference and scale photogrammetric models which are only relatively oriented.
The generation of the full three-dimensional buildings models has to pass through the absolute orientation phase which requires the knowledge of a certain number of distinguishable points on the model. For this purpose conspicuous points next to the lowest part of the roofs which are known from the procedure described in the point number 1. The precision achievable with this procedure of direct absolute orientation by means of GPS/INS system will be accurately investigated.

6. Application of the methods described on real cases.
The applications of described methodologies to practical situations will face up to many difficulties to be solved by using flexibility and by combining the different methodologies described above. Therefore, we will find applications cases close to ideal situations (for example a single building without obstructions all around) or situations where the field of view may be partially blocked by vegetation or gates. We will find other further complicated situations related to buildings in a historical centre where narrow roads interfere on the determination of camera’s exterior orientation parameters by GPS/INS system.

7. Calculation of volumes. The DTM and DSM data resulting from laser or aerial photogrammetry provide significant three-dimensional georeferenced information that can be used to obtain geometric attributes for territorial objects and, in particular, make possible a proper management of the volume of urban buildings. These volumes can be globally compared with those arising from cadastral maps and the from the knowledge of the heights referred to the eaves lines. It is very important for this purpose a correct filtering of ground points which determine the DTM, especially in the case of buildings located on land with a high slope where the presence of possible overhangs placed at ground level can produce problems to the automatic filtering of points cloud.