Feasibility study of stereo vision as a stand-alone sensor for obstacle detection.

Nolte, Armand (2017-03)

Thesis (MEng)--Stellenbosch University, 2017.

Thesis

ENGLISH ABSTRACT: The need for obstacle detection and the estimation of the relative distance of the objects is increasing in the field of robotics. The feasibility of using stereo vision as a means of detecting objects and their relative distance was investigated in this study. The aim was to implement obstacle detection within an area with moving objects and a moving vehicle. Objects were detected using disparity map generation. Different noise filtering techniques were applied to the disparity map to enhance the results. The noise filtering techniques tested for spatial aliasing, occlusion and low texture regions. To further improve the result, the disparity map of the current time frame was compared to that of its predecessor. The resulting disparity map showed large noise reductions but noise was still present. With slow-moving vehicles, the improved algorithm gives promising results.

AFRIKAANSE OPSOMMING: Die nut vir voorwerp opsporing en die bepaling van die relatiewe afstand van die voorwerpe word al hoe belangriker in die veld van robotika. Die haalbaarheid om stereo visie te gebruik as metode om voorwerpe op te spoor en die relatiewie posisie te bepaal, is in hierdie studie ondersoek. Die doel was om voorwerp opsporing in ’n gebied met bewegende voorwerpe en ’n bewegende voertuig toe te pas. Voorwerpe is geïdentifiseer deur ongelykheidskaarte te gebruik. Verskillende geraasfilterings tegnieke is op die ongelykheidskaarte toegepas om die resultate te verbeter. Die tegnieke het vir ruimtelike aliasering, afsluiting en lae-tekstuurareas getoets. Om die resultate verder te verbeter, het ons die huidige ongelykheidskaart met die vorige ongelykheidskaart vergelyk. Die gevolglike ongelykheidskaart het baie minder geraas bevat, maar die geraas was nog nie heeltemal verwyder nie. Met stadige voertuie gee die verbeterde algoritme belowende resultate.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/101168
This item appears in the following collections: