Using probabilistic graphical models to detect dynamic objects for mobile robots

Brink, Daniek (2016-12)

Thesis (DEng)--Stellenbosch University, 2016.

Thesis

ENGLISH ABSTRACT: An autonomous mobile robot must be able to identify moving objects in its environment, continually as it is operating, for accurate environment mapping and collision-free navigation. This is not an easy task, since most of what might be observed will appear to be moving due to the robot’s own motion. The task is further complicated by the inherent uncertainty in the pose estimates and environment measurements captured by the robot. In this work we focus on features in the environment whose 3D locations are measured over time, such as triangulated stereo image features. Our aim is to separate dynamic features from static ones and also to group the dynamic ones into separate objects. Existing approaches generally assume that the exact pose of the robot is known at every time step or, in order to estimate the robot pose, they assume that the environment is predominantly stationary. We avoid these assumptions through thoughtful consideration for the uncertainties involved. In order to model the uncertainties, as well as the statistical dependencies between observations and latent variables, we present a novel application of probabilisitic graphical models (PGMs) for dynamic object detection. Our PGM can be divided into two interacting components. The first relates to motion segmentation, in which all observed features are classified as static or dynamic, and the second relates to object segmentation, in which dynamic features on the same objects are clustered together. We also take care to accommodate for semi-static objects, which are objects that can be both stationary and dynamic during the observation period. Our design choices lead to a PGM containing both discrete and continuous variables. Tractable inference in such a hybrid model can be challenging, and we pay particular attention to this issue. It turns out that messages sent from continuous to discrete variables can be pre-computed, before loopy belief propagation is performed over the discrete variables. Experiments on the KITTI benchmark datasets indicate that our PGM approach performs well, and it often outperforms a state-of-the-art feature-based algorithm. We find that motion segmentation accuracy tends to improve as more observations of the same features become available, and that our method has the ability to handle semi-static objects successfully. The ability of our PGM to segment different objects is also seen to perform superior.

AFRIKAANSE OPSOMMING: ’n Outonome vry-bewegende robot moet oor die vermoë beskik om voortdurend bewegende voorwerpe wat in sy omgewing voorkom te identifiseer, ten einde akkurate kartering en botsing-vrye navigasie te bewerkstellig. Die taak word verder gekompliseer deur die inherente onsekerheid in die postuurafskattings en omgewingsmetings wat die robot neem. In hierdie proefskrif fokus ons op kenmerke in die omgewing waarvan die 3D posisies oor tyd gemeet word, soos verdriehoekte stereo-beeldkenmerke. Ons poog om dinamiese en statiese kenmerke van mekaar te skei asook om die dinamiese kenmerke in verskillende voorwerpe te groepeer. Bestaande benaderings aanvaar tipies dat die eksakte posisie en oriëntasie van die robot op elke tydstap bekend is of, in ’n poging om dit af te skat, word aanvaar dat die omgewing grotendeels staties is. Ons vermy hierdie aannames deur welbedagte oorwegings vir die betrokke onsekerhede. Om hierdie onsekerhede asook die statistiese afhanklikhede tussen waarnemings en latente veranderlikes te modelleer, stel ons ’n nuwe toepassing van grafiese waarskynlikheidsmodelle vir dinamiese voorwerpherkenning voor. Ons grafiese model kan in twee wisselwerkende komponente verdeel word. Die eerste komponent is verwant aan bewegingsegmentering, waarin alle waargenome kenmerke as staties of dinamies geklassifiseer word, terwyl die tweede komponent verwant is aan voorwerpsegmentering, waarin dinamiese kenmerke op dieselfde voorwerp saam gegroepeer word. Ons tref voorsorg om semi-statiese voorwerpe, wat kan beweeg en stilstaan in die waargenome tydperk, te akkommodeer. Ons ontwerpkeuses lei tot ’n grafiese model wat diskrete en kontinue veranderlikes bevat. Uitvoerbare inferensie in so ’n hibriede model kan uitdagend wees en ons skenk veral aandag aan hierdie kwessie. Dit blyk dat boodskappe wat van kontinue na diskrete veranderlikes gestuur word, vooraf bereken kan word, voordat vertroue-sirkulasie (“loopy belief propagation”) oor die diskrete veranderlikes uitgevoer kan word. Eksperimente op die KITTI maatstaf-datastelle dui aan dat ons grafiese model benadering goed presteer en dat dit gereeld ’n vooraanstaande kenmerk-gebaseerde algoritme uitstof. Ons bevind dat die akkuraatheid van bewegingsegmentering neig om toe te neem namate meer waarnemings van dieselfde kenmerke beskikbaar word en dat ons metode die vermoë het om semi-statiese voorwerpe suksesvol te hanteer. Die vermoë van ons grafiese model om verskillende voorwerpe te segmenteer blyk ook beter te vaar.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/100224
This item appears in the following collections: