With the use of Building Information Modeling (BIM), digitalization has been gaining ground in the construction industry in recent years. However, the use of digital technologies is largely limited to the planning phase. The extension of the methods currently used in planning to all life cycle phases, including construction and facility management, offers enormous potential for cost and resource savings for the entire construction and real estate industry.
Therefore, an innovative system should be developed to quickly and easily record the actual state of construction and installation services (as-built) and to automatically compare it with the target state of planning in the form of a BIM model, using novel intelligent methods. Such a system supports on the one hand the documentation and control during the construction phase and on the other hand the facility management during the operational phase of the building by means of digitally recording existing buildings. For this system, new methods for the recording of the actual state by means of mobile optical sensor systems, location of these sensor systems indoors, mechanical recognition and characterization of building components by means of deep learning algorithms, and for the automatic comparison of actual and target state, are developed and extended. Here we want to evaluate methods and the necessary technological basics.
Multiple tasks are available and can be worked on individually or in parallel depending on PR/BA/DA:
Explore a LIDAR/Depth sensor based SLAM tracking method developed by KUDAN in the context of the construction industry and integrate the method in a real-time scanning system.
Explore the KinectFusion and related algorithms and create 3D reconstructions of the real environment using RGB-D sensors or LIDAR.
Improve an existing 3D reconstruction implementation by using prior knowledge, such as an approximate 3D model of the environment (BIM-Model) and the precise position of the sensor (from the abovementioned accurate SLAM-Solution). This prior knowledge should allow more precise 3D reconstruction with an algorithm that requires less memory for the computations.
Explore / Implement methods to match a 3D model with a point-cloud
Explore / Implement methods to detect/segment geometric primitives in a point-cloud
Knowledge of English language (source code comments and final report should be in English)
Knowledge of C++, computer vision and related libraries like OpenCV is advantageous
The project should be implemented in C++ as a stand-alone application, based on libraries like OpenCV and PCL (Point Cloud Library) ideally platform independent (Linux, Windows). Depending on the sub-task for Visualization also Unity and C# could be used.
A bonus of €500/€1000 if completed to satisfaction within an agreed time-frame (PR/BA or DA)