tailieunhanh - Innovations in Robot Mobility and Control - Srikanta Patnaik et al (Eds) Part 8

Tham khảo tài liệu 'innovations in robot mobility and control - srikanta patnaik et al (eds) part 8', kỹ thuật - công nghệ, cơ khí - chế tạo máy phục vụ nhu cầu học tập, nghiên cứu và làm việc hiệu quả | 3 Multi View and Multi Scale Image Based Visual Servo For Micromanipulation 129 robust feature detection with pixels noise level even with occlusion see Fig. and Fig. . Fig. . Estimation of features ellipse and centroid detection Fig. . A selection of close-ups illustrating degrading focus and contrast as well as occlusion To estimate motion from images the motion estimation performance from successive frames of images is tested. Fig. shows the comparative results among performance of optical flow OF and Markov Random Fields MRF . The root-mean-square RMS error is compared. From the results the MRF method shows good estimation accuracy even for long displacement compared to the optical flow method. The experiment result shows that after 10 frames the estimation error of optical flow increases sharply that is because the optical flow based algorithms extract a dense velocity field from an image sequence assuming that image intensity is conserved during the displacement. This conservation law does not hold when the error of distance estimation between images accumulates and the distance becomes large. Successive frames of the same scene are typically views of the same objects whose images are shifted in the frames due to the motion. By MRF we can assess that displacement despite 130 R. Devanathan et al. very different local frame pixel values after large motion since such frames are still similar. The statistical models of MRF characterize images and allow computations of distances yet are relatively insensitive to translation. In fact MRF relates the spatial and temporal information together to find the most likely displacement between image frames. Conclusion The experiments show the validity of the proposed method. By using both macro view and micro view at the same time the proposed method appears to be good at solving some of the problems caused by uncertainty in micromanipulation system as demonstrated by the experiments. This .