Hand-eye Robotic Visual Servoing Around Moving Objects Using Active Deformable Models

with a controolkd camera. Our .ppmrch is based on computer-generated ddomuble modrk, w h i i move and deform in order to adhere IO thc contourn of dK object king ttackd. A ddotmable or elastic d e l K a mesh of Urifici.l massive paticks coIu).Qcd to each &r by artificial elastic springs. Eachputick intcrwo with the object through attracting forcca c r u d by high valuer in the image-gdicni map. The elastic model moves and deforms in the image pluw wxordmg to the laws of c k i l mechanics. Our deformabk model. ue similu to, yet quik di5mnc than, the ucfive d e s i n d u c e d in c o m p u y ~ gnph ia by Tenopoubs [I]. The dirplvcmcnts and deformations of rhe elastic model ue directly p'occsKd to estimate the motion of the endefiector. which will visually main& the model's initial position and shape. The popoleddefanubk models ue yvy robust with respect to the shape of the ogersd objecq which am q u i r # l to be rigid bod= with wcll&fined bouniUiam*nnrofimagegrdienL nKmainprobkmofthevisualtracking task ktocanpukdu .ppropri.Ucamcnmotioa from the position andshape of the elvtic m0dcL a d d k c @ r ~ optimal filming techniqueshave also been r u c w f u l l y implanentsdinrul-rime. The ICSC of this paper ddrrsper the following: Section 2 briefly reviews previour wortrr dut ddrcrc the issue of robotic visual rrvoing; Section 3 g k s 8 dctded daaiptioa of the defornubk model: Section 4 provides a p o o f of the Uabilii of the etrrtic model& as wcu as the methods uscd to sckctdynmicpuunctcn: Section 5 presents the p d u r e that estimates the desk4 motion of the cnd-cffector. Section 6 gives an overview of our robotic codguntionandimage pnxxssing hardwue; Section 7 prtscntsexperimenul result. of srvoingthe endcficuor around an object moving in a plane paralkl to the haage pluw (3-WF visual servoing). In an urcmpc to PIIOoth the rcquenoc of computed