View source for Robust data association for high-speed SLAM

== '''Part 1: project profile''' ==

=== Project name ===

Robust data association for high-speed SLAM

=== Project short description ===

The purpose of this project is to approach the problem of robust data association in the context of Simultaneous Localization And Mapping (SLAM) for mobile robots using digital cameras as input data sources, by finding a method that could allow the SLAM problem to be solved very quickly, possibly even in real-time, thanks also to the use of parallelization strategies.
In particular, a study of the possible applications of parallelization strategies inspired by the Particle Swarm Metaheuristic will be performed.

=== Dates ===
Start date: 2009/07/22

End date: Not already defined

=== Website(s) ===

None

=== People involved ===

===== Project head(s) =====

[[User:MatteoMatteucci|Prof. Matteo Matteucci]]

===== Other Politecnico di Milano people =====

[[User:DavideMigliore|Davide Migliore]]

===== Students currently working on the project =====

[[User:FilippoRottoli|Filippo Rottoli]]

===== Students who worked on the project in the past =====

None

===== External personnel =====

None

=== Laboratory work and risk analysis ===

The work will consists only of software development, test and analysis, so no risks are associated with the laboratory's activities for this project.


== '''Part 2: project description''' ==

==== Premises ====

A problem is defined as a Simultaneous Localization And Mapping (SLAM) one if it asks to a mobile robot equipped with a sensor to build a map of the environment in which it is placed and, at the same time, to use this map to compute its location.

The strictly coupled data association problem consists of relating sensor's observations with elements in the map by finding the correct matches.

This project has as its final goal to detect a robust data association algorithm for the SLAM problem involving mobile robots using digital cameras as sensors, that could be executed at high speed, possibly even in real-time, thanks also to the use of parallelization strategies.
At this regard, the introduction of techniques inspired by the Partice Swarm Metaheuristic will be considered with great attention, due to both the high potential in the distribution of the execution effort of such techniques, and to the good results that Particle Swarm Optimizers have demonstrated to obtain (at least on average), also on highly complex problems.

==== Oganization of the work ====

This project will be organized according to the following main steps:
# At first, a set of robust and efficient data association algorithms (e.g.: JCBB, RJB, RANSAC, Preemptive RANSAC, Davison and Kli Mutual Information Algorithm, MLSAC and SwarmSAC), selected from the state of the art, will be implemented and compared, producing a first batch of benchmark results.
# Then, the benchmark data will be analyzed in order to detect the best algorithms and the possible key features to calibrate in order to obtain better performances without sacrificing the results' quality and the robustness.
# A further step (as stated in the previous section) will consists in exploring the possibility to insert techniques inspired by the Particle Swarm Metaheuristic within the selected algorithms (or modifications of those already used, such as for the case of SwarmSAC), due to the high level of parallelism that characterizes this kind of Metaheuristic.
# Eventually, comparisons between all the algorithms implemented in the project will be performed and final conclusions will be drawn.

==== Technical details ====

All the project's algorithms will be written in C and C++ code using:

* OpenCV libraries because of the fact they are becoming the "de facto" standard in the realization of Computer Vision applications and due to the fact they are optimzed to run efficiently on the platforms for which they are available.
* Intel IPP and Intel Math Kernel libraries in order to build applications that could take advantage of high-performance parallel execution especially on multi-core machines.

==== Main references ====

* Hartely R., Zisserman A., ''Multiple View Geometry in Computer Vision, 2nd ed.'', Cambridge University Press, 2004
* Bradki G., Kaehler A., ''Learning OpenCV: Computer Vision with the OpenCV Library, 1st ed.'', O'Reilly Media Inc., 2008
* Engelbrecht A. P., ''Fundamental of Computational Swarm Intelligence, 1st ed.'', Wiley, 2006
* Poli R., Kennedy J., Blackwell T., Freitas A., ''Particle Swarm: The Second Decade'', Hindawi Publishing Corp US SR, 2008
* Eberhart R. C., Shi Y., Kennedy J., ''Swarm Intelligence, 1st ed.'', Morgan Kaufmann, 2001