| VISUAL SENSING AND INTERPRETATION: IMAGE UNDERSTANDING POINT OF VIEW (2007) | |||||||||||||
Abstract | |||||||||||||
| Vi\u.tl wnvng and inrcrprctation of cn\ironmcnts are widely rccogniicd as an importrlnt capability for robots to bchnbc intcllifcntly and flexibly. Yet. it seems that succcssful applications of \ision in robotics arc liinitcd to such usks as inspecting and locating objects, and also that the mcthods emplo)cd arc mo,tly ud hi> ( and tuned to thc parricular usks bj cdrcful cnginccring. Partly bccausc vision is so difficult. and parrl) bccausc c\cn thc chcdpcst solutions can still be so uscfiil. approaches to vision problcms haw had a tcndcncy to bc ad hoc and hcuristic Ilcccntly. howc\cr. ncw thrusts in computcr \ision arc cmcrginp. most notabl) in thc Imagc bnderstiinding community. that tr) to pursue more systematic and computational approachcs. [l] This papcr attempts to introduce such Imagc Undcrstand~ng approachcs to vision. It first prcsents the au[hor's \ icv. of structure of \ ision and a brief diwussion of current succcssful iipproachcs in robotics \ision in tcrms of thc pi-oposcd \icw. The following thrcc issues are idcntificd as tlic important ones to crcatc gcncral robotic 1 i\ion which can scc and intcrprct thc cnvironmcnts. 0 l~~onn;rli/c thc phvsical constraints for rcasoning the sccnc fcaturcs from thc imagc fcaturcs. 0 Modcl tlic ta,k domain in a gcncric modcl and usc it to andyic images. 0 Construct thc sccnc description in a bottom-up manncr Tlicn. rcprc\cnt;i~i\e rcscarch progrcs \ in lmagc C!ndcrst;lnding I \ rc\icwcd which addrcsscs tbcsc issues. | |||||||||||||
Publication details | |||||||||||||
| |||||||||||||