With the increase in the amount of data one can access comes the need to have more efficient ways to retrieve the very same data. With the advent of the internet the data one is able to access increased exponentially and still continues to increase at the same pace. Conventional data like documents are comparatively easy to index and processing the same is less intensive. But processing 3D models is a complex task, quite simply because of the way they are stored. They, unlike text documents contain co-ordinates and texture information. They need a different approach to be indexed and to be searched accurately, efficiently and intuitively. The internet is full of thousands if not possibly millions of three dimensional object models. They are the next step in information retrieval systems and search engines. They are the successors to two dimensional pictures. Assume that while shopping one could also rotate and view the product from all angles. Most online sellers already provide such facilities but finding a model of a three dimensional object is a shot in the dark with a search of the files available. What if one did not know the name of the object? What if they just had a model and wanted to find similar models? This task would involve the search and retrieval of 3d models from a large dataset, that too with the ability to show similar shaped models, to to be able to match all models belonging to the specific class or type of the query model, not to mention being able to match objects invariant of deformations and similarity transformations of rotation, scale and translation. Shouldn’t there be a way to look through all possible matches and show us the results. The underlying 3d object detection technology can be applied to many fields including robot navigation, industrial manufacturing, automated driving, robotic surgery, and for recognizing human hand gestures for an improved HCI.
About our project
With the advancement of information technology in our society, we can expect that computer systems to a larger extent to be embedded into our environment. Our dependence on search engines for various tasks is already well established. The internet, over the years, has amassed huge amounts of data that needs to be indexed for efficient retrieval. Currently search engines are capable of indexing only limited types of data, but as our dependence on the internet grows, as the processing power of our desktops increases, so will the types of data we process upon. And thus the need to index the more complex types of data. The internet contains millions of detailed 3D models in a huge category of objects. Searching according to their filenames or through references of the pages they appear in, are at present the only way to search for them, there exists no automated way for comparing one 3D model with a search set and come up with a set of possible matches. Our aim was to develop a reusable engine to be able to search and index 3D meshes. To be able render the possible results and the query model itself. The indexing of 3D models is such a precarious task for the simple reason that the search not only has to be resolution independent as in case of search for two dimensional pictures but it also has to be independent of translation, orientation and distortions. The approach must not only be accurate but also be quick to compute, invariant to transforms, the indexed data must also be condensed enough to be stored.
The problem is to create a 3d model search engine which provides an intuitive query interface for a user to search a query model in a large database of indexed 3d objects both accurately and efficiently. The search engine must devise and implement a shape matching algorithm which is capable of matching shapes with scale, translation, rotation and deformation invariance. The query interface must allow a user to import a 3d model file, render and view the model (ability to rotate and scale the model), and then search the model database for similar models. The user should be able to adjust the engine’s settings to tune the search results’ speed according to the computing resources on the client machine. The interface must also allow search results to be rendered and saved. The application must also have Model Database Manager allowing import, delete, rename, move and view options for models and allowing addition, deletion and renaming of categories. All the models in the database must be indexed with their respective shape descriptors. The search engine should use its own optimized file format for 3d object representation and feature reading and writing model data from the same. We formulate this problem as an model database indexing problem, where the closest matches for a query 3D model are retrieved from a large database of indexed 3D models. The problem is to find the most similar images in a large database of 3D models, given an input query model. The shape descriptor parameters of the retrieved models are used as estimates for the comparison with query 3D model. Each model is labeled with parameters describing its shape parameters. Estimation of 3D models is useful in many applications, ranging from the 3D object search to human-computer interaction to automated recognition of sign languages.
A) 3D Model Representation An optimized file format for database storage of 3d models Shape descriptor calculation and storage B) 3D Model Search Intuitive search query interface Importing a 3d model in popular file formats Generation of points on model surface avoid shape ambiguity in low resolution models Calculation of shape descriptor for efficient and accurate searching and indexing Matching of shape descriptor for retrieving search results in the order of similarity Internal Model Viewer for rendering search results C) Tools Model Viewer for Rendering the 3d model (rotate and scale UI) Database Manager (import, move, rename, delete, view, manage categories) Shape Distribution Calculator EMD Calculator Point Generator Screenshots
Shape Distribution Calculator
These details are provided for information only. No information here is legal advice and should not be used as such.