Page - 75 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

(a)LFDepthTail (b)Combined Depth Tail (c) GT (d)LF DepthHead (e)CombinedDepthHead (f) GT Figure 4: Combination of light field and photometric stereo in order to improve fine structures in the light field depth estimation. Both rows show a part of the tested object: (a),(d) original light field depth, (b),(e) improveddepth by photometric stereo, (c),(f) ground truth depthmap. 4. ExperimentalResults We performed experiments on several coins, using the multi-line-scan setup, an example is shown in Fig. 5. Transporting the coin, shown in Fig. 5a, along the conveyor belt, we acquired light field and photometric stereo data. The depth information gained from light field data is shown in Fig. 5b. Surface normals are estimated from the same photometric light field data through the detection of a specularlobeineachimagelocation. Thespecularlobepositioncorrespondswiththelocalorientation of thesurface. TheimagedepthiscalculatedusingthesurfacenormalsasdescribedinSubsection3.2. In Fig. 5c we obtained a refined solution by the combination of photometric stereo depth and light fielddepth, asdescribed in Sec. 3. 5. ConclusionandDiscussion Wediscussed theprosandconsofpassive (lightfield)andactive (photometric) stereoapproaches for depth estimation on virtually generated data. A way to combine both approaches in order to achieve a more precise depth estimation was presented. We also showed initial results on our multi-line- scan setup, which is acquiring light field data and photometrically varying data at the same time, i.e. while observing the object under relative motion. Relative motion between the object and the acquisition device is a typical configuration in industrial vision systems, thus this setup fits well for suchapplications. Initial resultsweregivenandthesuggestedcombinationschemewasdemonstrated. Further work will cover a more complete evaluation, as well as an improvement of the combination scheme. References [1] R. C. Bolles, H. H. Baker, and D. H. Marimont. Epipolarplane image analysis: an approach to determiningstructure frommotion. Int. J.Comp.Vis., 1(1):7–55,1987. [2] R. T. Frankot and R. Chellappa. A method for enforcing integrability in shape from shading algorithms. IEEETrans. Pat.Anal. andMach. Intell., 10:439–451,1988. 75