|M.Sc Student||Tansky Dmitry|
|Subject||Data Fusion and 3D Geometric Modeling for Multi-Scale|
|Department||Department of Mechanical Engineering||Supervisor||Professor Anath Fischer|
|Full Thesis text|
Inspection analysis of 3D objects has progressed significantly due to the evolution of advanced sensors. State-of-the-art sensors facilitate surface scanning both at the macro and the micro levels. Currently the main challenge is to integrate this multi-scale diverse scanned data into a single geometric model. In the inspection field, data from multi-scale sensors offers significant advantages over single-scale data. Most data fusion methods are single-scale and are not suitable in their current form for multi-scale sensors.
In this research, two new generic frameworks are proposed for data fusion from multi-scale sensors: (a) Single-Level Multi-Sensor (SLMS) framework and (b) Hierarchical Multi-Sensor (HMS) framework. These frameworks utilize the multi-scale properties of the sensor scanning and incorporate some common framework models from the literature. The SLMS framework is relevant for cases in which all multi-scale sensor data is merged in one level, while with the HMS framework data from multi-sensors can be added hierarchically.
For robust performance, both the SLMS and the HMS data fusion frameworks require a common working core that is robust and flexible and can be used for a variety of sensors. The proposed core includes five stages: (a) Data acquisition from multi-scale (CMM and laser) sensors; (b) Mesh reconstruction from sampled data; (c) Feature detection; (d) Feature matching; and (e) Data merging into a multi-scale model;
Between the main characteristics and advantages of the proposed SMLS and HLS frameworks are: (a) Different inspection technologies can use these frameworks; (b) The number and type of sensors are not limited; (c) Diverse data sets can be added adaptively; (d) Multi-scale data sets can be merged into a single multi-scale model; (e) These frameworks are using common working core; and (f) the approach is not limited to closed meshes only, but can also handle open meshes;
A novel approaches for merging of multi-scale model within the frameworks are proposed: (a) Selective merging based on the new error map principle; (b) Retentive merging of low resolution dense data with high resolution sparse data, using the statistical GP approach; and (c) using the new B-spline correction function;
The feasibility of the proposed frameworks are demonstrated on multi-scale models, generated by 2.5D scanned surfaces from CMM and laser scanner and on 3D multi-scale synthetic data from CAD models, when synthetic models include noise appropriate to real sensor measurements. The evaluation of different parameters for B-spline correction function provides significant reduction of error values of data.