1. Integration of IoT sensors to 3D models

2. Update BIM models using point cloud

3. Indoor space subdivision for navigation applications

4. Simplification of indoor features in 3D BIM models

5. Point cloud classification for indoor environments

6. Voxelisation of objects and 3D analysis in Unity3D

7. Visibility analysis for open spaces

8. Navigation using 3D space-based navigation model

9. Automatic 2D areas subdivision (e.g., convex polygons)

10. Creating non-overlapping coverages of objects from 2D unstructured data (e.g., CAD file)

11. Trajectory-driven influential navigation panel placement

12. Reconstructing 3D city models from historical image data

13. Urban Point-of-Interest mining

14. An interactive visualization system to analyse and predict built environment dynamics

15. Flightpath for navigating drone indoors

16. Following pre-computed flight path indoors

17. Procedure for optimising the process of indoor scanning

18. Developing a voxel database schema

 

1. Integration of IoT sensors to 3D models

Description: Internet of things (IoT) sensors occupy a central place in the concept of Smart Cities and Smart buildings. Most of them have strategical locations, allowing them to collect data that often relates to the spatial use of the indoor environment. In this research, the goal is to investigate the best way of integrating IoT sensors and 3D indoor models of different granularity to manage them in one hand and visualize their output on the other hand.

Tasks: Develop a spatial database schema for integrated data management and interfaces for visualisation.

Recommended skills: Familiarity with spatial data (geometry, positioning, etc.), some experience with coding (any language, any level), experience with DBMS (e.g. PostgreSQL, PostGIS, etc.) also appreciated.

Contact: Abdoulaye Diakite (a.diakite@unsw.edu.au), GRID, BE UNSW

2. Update BIM models using point cloud

Description: Building Information Models (BIM) provide a detailed 3D representation of buildings. However, they describe the building as-designed (AD) and commonly differ from the building as-built (AB). Thanks to advances in surveying and remote sensing, new tools such as laser scanners allow producing precise 3D up-to-date point clouds of existing construction. The challenge is then to compare AD models and AB point cloud data and update AD models so as to keep the information about the building always up to date.

Tasks: Develop algorithms for matching point clouds and BIM models to detect changes and update the geometry of BIM models.

Recommended skills: Notions in GIS (vector and raster geometry, topology, semantic, etc.), some experience with coding (any language, any level), experience with DBMS (e.g. PostgreSQL, PostGIS, etc.) also appreciated.

Contact: Abdoulaye Diakite (a.diakite@unsw.edu.au), GRID, BE UNSW

3. Indoor space subdivision for navigation applications

Description: One of the main challenges in providing tailored indoor navigation is to be able to describe the indoor space in a way that fits to the purpose. The theory behind navigation networks abstracts every space as a node and every space adjacency as an edge. Therefore, if an indoor space is not subdivided properly, its detailed information will not be reflected in the network. For example, if one room is one node of a network, an agent can be guided in the best case to the room, while if that room is subdivided into several sub-parts, then the agent can be guided up to those sub-parts. In this research, the goal of the investigation is to identify strong and consistent criteria on which one can rely to automatically subdivide indoor spaces.

Tasks: Developing algorithms for space subdivision, as space is represented as vector volumetric model.

Recommended skills: Notions in GIS (vector and raster geometry, topology, semantic, etc.), some experience with coding (any language, any level), experience with DBMS (e.g. PostgreSQL, PostGIS, etc.) also appreciated.

Contact: Abdoulaye Diakite (a.diakite@unsw.edu.au), GRID, BE UNSW

4. Simplification of indoor features in 3D BIM models

Description: Building Information Models (BIM) provide a detailed 3D representation of buildings, including what they contain (e.g. furniture, equipment, etc.). Recently, the flexible space subdivision (FSS) framework for indoor navigation requires considering those contained objects as features occupying a non-navigable space (O-Spaces). Thus, they must be excluded from the space used for navigation. The problem is that such indoor features are very complex in geometry and need to be simplified. The
challenge of such geometric simplification is to maximize the shape simplicity while minimizing its storage occupancy. For this purpose, the investigator can explore several kinds of 3D representations (polyhedron, octree, voxel, etc.).

Tasks: Developing algorithms for aggregation and simplification of 3D objects

Recommended skills: Notions in GIS (vector and raster geometry, topology, semantic, etc.), some experience with coding (any language, any level), experience with DBMS (e.g. PostgreSQL, PostGIS, etc.) also appreciated.

Contact: Abdoulaye Diakite (a.diakite@unsw.edu.au), GRID, BE UNSW

5. Point cloud classification for indoor environments

Description: Point clouds of indoor environments are captured on a daily basis. However, they are mainly used for simple measurements without extracting more information from raw data. Classification has the purpose to determine to which object each point belongs. In indoor environments there are various types of objects (walls, ground, ceiling, stairs, furniture, etc.). Thus, the main goal is to propose a methodology which can automatically classify all point clouds into objects. A possible technique for their classification is through using voxels, which represent point clouds in more structured way.

Tasks: Transformation of point clouds into voxels. Classification of voxels (point clouds) to present different objects in indoor environments.

Recommended skills: Programming language such as Python (numpy and Pandas lib) or J. To visualise the voxels Unity3D would be used.

Contact: Mitko Aleksandrov (mitko.aleksandrov@unsw.edu.au), GRID, BE UNSW

6. Voxelisation of objects and 3D analysis in Unity3D

Description: Voxels have been used in many fields (medicine, gaming, architecture, urban planning etc.) to represent 3D objects as collections of cubes. For example, Minecraft uses voxel-based structures to visualise 3D environments. The main benefit of transferring vector objects into voxels is
to easily perform different 3D analysis. The goal of this project is to identify a quick way of transferring 3D vector objects into voxel representation. A possible solution is to first split vector objects into chunks in order to easily perform multiprocessing techniques.

Tasks: Chunking vector data into smaller 3D areas. Voxelisation of vector data. Querying voxels.

Recommended skills: Understanding of game engines and programming languages. Unity 3D is the suggested platform along with C# programming language.

Contact: Mitko Aleksandrov (mitko.aleksandrov@unsw.edu.au), GRID, BE UNSW

7. Visibility analysis for open spaces

Description: For many applications the field of view and related space analysis are of critical importance. For example, knowing the visibility within indoor and outdoor spaces, many guiding systems, advertisements and CCTVs can be positioned in a better way. The main goal of this project is to identify which spaces and objects are more visible during specific times of day. The trajectory of moving people and the spatial representation of 3D environments will be used as input data. A possible solution is to use raycasting techniques to model the visibility of people within 3D environments.

Tasks: Import of 3D environments in Unity. Modelling visibility area of people. Identifying visible indoor and outdoor spaces.

Recommended skills: Familiarity with game engines and programming languages. Unity 3D is suggested platform along with C# programming language.

Contact: Mitko Aleksandrov (mitko.aleksandrov@unsw.edu.au), GRID, BE UNSW

8. Navigation using 3D space-based navigation model

Description: Path computation algorithms require a network. One of the popular approaches to derive a network automatically is by leveraging 3D spaces and applying the 3D Poincare Duality theory. That is, a 3D space model is transformed to a network such that a space is represented as a node, and the face shared between two spaces is treated as an edge. Having the navigation network, different navigation algorithms can be used to perform path computation. This project concentrates on two tasks: 1) Leveraging 3D spaces from commonly available maps and models to derive navigation networks for different navigation modes (walk, drive, fly) and 2) Apply navigation algorithms (e.g., Dijkstra, Ant, A*) to perform route computation for different users

Tasks: Develop algorithms for automatic network derivation.

Recommended skills: Familiarity with programming languages. Rhinoceros 3D + Grasshopper + Python script is preferable.

Contact: Jinjin Yan (jinjin.yan@unsw.edu.au ), GRID, BE UNSW

9. Automatic 2D areas subdivision (e.g., convex polygons)

Description: In many cases large and long areas, such as long corridors, big squares, need to be subdivided in smaller parts because of semantic identification (e.g. waiting area) or for the purpose of
more accurate localisation. Therefore, before deriving a network (using Poincare Duality theory), such large areas should be subdivided into subspaces based on certain criteria (e.g., convexity, distances between walls, visibility, etc.).

Tasks: Develop algorithms for automatic 2D space subdivision and network derivation.

Recommended skills: Familiarity with programming languages. Rhinoceros 3D + Grasshopper + Python script is preferable.

Contact: Jinjin Yan (jinjin.yan@unsw.edu.au ), GRID, BE UNSW

10. Creating non-overlapping coverages of objects from 2D unstructured data (e.g., CAD file)

Description: 2D blueprints are commonly available as a two-dimensional data source. In general, these blueprints are drawn by urban planners, designers or surveyors for some specific applications, such as visualization of the whole area, guiding constructions, or 2D mapping. Such maps often consist of disconnected lines (spaghetti), which represent different objects and therefore they are not readily applicable for path computation. For instance, if a footprint of an object is not an enclosed polygon, this footprint cannot be associated with a navigation node in a network. Therefore, before using these 2D blueprints for navigation, the first step is defining footprints of objects as polygons.

Tasks: Develop algorithms for automatic creation of 2D polygons from 2D CAD blueprints.

Recommended skills: Familiarity with programming languages. QGIS, Rhinoceros 3D + Grasshopper + Python script is preferable.

Contact: Jinjin Yan (jinjin.yan@unsw.edu.au ), GRID, BE UNSW

11. Trajectory-driven influential navigation panel placement

Description: Navigation panels play an important role in urban navigation. Moreover, it literally drives people to search, interact, and transact. Currently, existing navigation panel placement only leverages
important locations or intersections. Such a straightforward approach often leads to coarse-grained performance estimations and undesirable navigation placement plans. To enable more effective placement strategies, we are interested in developing a fine-grained approach by leveraging the user/vehicle trajectory data.

Tasks: Develop algorithms for appropriate navigation pane placement using the trajectory of users (pedestrians or cars).

Recommended skills: Familiarity with programming languages, understanding of GIS and navigation

Contact: Wei Li (wei.li@unsw.edu.au), GRID, BE UNSW

12. Reconstructing 3D city models from historical image data

Description: Automatic generation of 3D city models attracts significant interest. Also, people miss the old built environment and want to reconstruct the city model through the historical image data. Working with historical image data is substantially more difficult, as there are significantly fewer images available and the details of the camera parameters which captured the images are unknown. This project will concentrate on automatic generation of 3D city models from historical images of cities to expose historical data to users through an immersive walkthrough experience.

Key words: Architecture (buildings), Human computer interaction (HCI), 3D modeling, computer vision

Tasks: Develop algorithms for 3D reconstruction from images

Recommended skills: Architecture (buildings), Human computer interaction (HCI), 3D modeling, computer vision

Contact: Wei Li (wei.li@unsw.edu.au), GRID, BE UNSW

13. Urban Point-of-Interest mining

Description: Cities are getting more complex, busy and dynamic and therefore dedicated Location-Based Services (LBSs) are needed to effectively improve smart urban living. One of the topics of interest in this direction is to investigate the Points-Of-Interest (POIs) where people may be interested in and make them apparent. This project will concentrate on the problem of mining POIs from the collections of users’ social network (for example Facebook, Google, Instagram).

Tasks: Develop algorithms for data mining of PoI from Social Networks

Recommended skills: The candidate should have an understanding of Urban Computing, Data Mining, Location-Based Social Networks

Contact: Wei Li (wei.li@unsw.edu.au), GRID, BE UNSW

14. An interactive visualization system to analyse and predict built environment dynamics

Description: In this work, we aim to develop an analytic and visualization system for stakeholders to understand, track, and predict the built environment dynamics in an urban area. First, we provide an interactive data visualization interface that allows various users to easily investigate the relationships among different construction types, geographical regions of interest, and the duration of construction over time. Then, a regression-based prediction model is devised to forecast the future development for a certain construction type of interest and a specific region of interest.

Tasks: Develop appropriate system architecture and prediction model

Recommended skills: Interactive visualization, urban computing, regional analysis, clustering, regression

Contact: Wei Li (wei.li@unsw.edu.au), GRID, BE UNSW

15. Flightpath for navigating drone indoors

Description: Maintaining up-to-date 3D indoor models of public buildings is becoming increasingly important. A large number of scanning devices and platforms are currently available but most of them require manual operation. This research concentrates on automatic recording indoor environments using small, low-cost drones (such as quadcopter Mavic). Similar to outdoor data collection with drones, a flight path is needed to guide the drone through the indoor environment. The aim of this research is to investigate indoor conditions such as physical constraints, light conditions and texture patterns that influence the quality of the video images and therefore impact the flight path.

Tasks: Develop and test strategy for indoor data collection with Mavic quadcopter

Recommended skills: Understanding of Photogrammetry, Image processing, Point cloud processing. Software to be used for 3D reconstruction (Pix4D/ Agisoft)

Contact: Sisi Zlatanova (s.zlatanova@unsw.edu.au ), GRID, BE UNSW

16. Following pre-computed flight path indoors

Description: Drones steadily find their daily place for outdoor guidance and delivery (https://www.youtube.com/watch?v=prhDrfUgpB0) and are waiting for their use indoors. This project is concentrating on programming a drone to follow a predefined path (without GPS positioning) to a specific location. Resent research and developments have resulted in many algorithms that can provide a safe flight path taking into consideration physical obstacles in a building. The next step is to program a drone to follow the path.

Tasks: Program a drone (e.g. Mavic) to follow a precomputed path.

Recommended skills: The candidate should have strong skill in robotics, Indoor Navigation, Electrical engineering

Contact: Sisi Zlatanova (s.zlatanova@unsw.edu.au ), GRID, BE UNSW

17. Procedure for optimising the process of indoor scanning

Description: Indoor environments are usually very complex for scanning and normally take significant time and effort to collect data from all corners. This research aims at developing an optimisation procedure for selecting either the stations for laser scanner or computing the trajectory of a mobile scanner. A possible solution is estimating the field of view from a scanner location, using ray tracing.

Tasks: Develop an algorithm for optimising locations or trajectory of a scanner. Zeb CAM is available for the research.

Recommended skills: Programming, Optimisation algorithms, Indoor Navigation

Contact: Sisi Zlatanova (s.zlatanova@unsw.edu.au ), GRID, BE UNSW

18. Developing a voxel database schema

Description: Voxels are beneficial for 3D modelling as they provide a unified data structure for integrated maintenance of indoor, outdoor, above and below surface objects and therefore very appropriate for environments 3D spatial analysis and simulations. However, voxel models can become very large in size especially if whole precinct and cities are envisaged. This research concentrates on a database schema able to maintain multi-resolution voxel representations and their properties.

Tasks: Develop a database schema and SQL scrips for import and management voxels. PostGIS is recommended for use.

Recommended skills: Knowledge in spatial DBMS, SQL, Programming, semantics, geometry

Contact: Sisi Zlatanova (s.zlatanova@unsw.edu.au ), GRID, BE UNSW