Academia.eduAcademia.edu

Figure 10 – uploaded by afaq ali

See full PDFdownloadDownload figure
all the objects were correctly recognized and segmented from the scene. We also performed a quantitative evaluation of the proposed technique on the

Figure 10 all the objects were correctly recognized and segmented from the scene. We also performed a quantitative evaluation of the proposed technique on the

Related Figures (14)

S003 1-3203(16)30342-9 http://dx.doi.org/10.1016/j.patcog.2016.10.028 PR5936
S003 1-3203(16)30342-9 http://dx.doi.org/10.1016/j.patcog.2016.10.028 PR5936
Figure 2: Illustration of the proposed Keypoint based Surface Representation (KSR) tech- nique. The colored lines show the distance measured between the keypoint 7 and the other detected keypoints. The two nearest neighboring keypoints of the F are selected based on their minimum distance (shown in blue) to #. The same procedure is repeated for keypoint Pi+1 Figure 1: Proposed methodology for computing Keypoints based Surface Representation (KSR).
Figure 2: Illustration of the proposed Keypoint based Surface Representation (KSR) tech- nique. The colored lines show the distance measured between the keypoint 7 and the other detected keypoints. The two nearest neighboring keypoints of the F are selected based on their minimum distance (shown in blue) to #. The same procedure is repeated for keypoint Pi+1 Figure 1: Proposed methodology for computing Keypoints based Surface Representation (KSR).
Figure 3: Recall vs 1-precision curves. Performance of KSR with state-of-the-art keypoint detectors.
Figure 3: Recall vs 1-precision curves. Performance of KSR with state-of-the-art keypoint detectors.
performance with HKS followed by MeshDoG and DoG (Fig. 4(c)). As the noise
performance with HKS followed by MeshDoG and DoG (Fig. 4(c)). As the noise
with a high level of noise.
with a high level of noise.
Figure 6: Feature matching and Regsitration. (a) Feature matching between two views. Only few feature correspondences are shown. (b) Registered views. This Figure is best viewed in color.
Figure 6: Feature matching and Regsitration. (a) Feature matching between two views. Only few feature correspondences are shown. (b) Registered views. This Figure is best viewed in color.
Figure 7: Example to illustrate the results of our multiview range image registration. (a) Initial set of 11 range images of the Chef model. (b) Coarse registration result by KSR based method. The different views are shown in different colors. (c) The coarse regis- tration result shown in one color for visual clarity. (d) The final 3D model in the UWA Dataset (Ground truth). Figure best viewed in color. Performance evaluation of the proposed technique against popular spin image:
Figure 7: Example to illustrate the results of our multiview range image registration. (a) Initial set of 11 range images of the Chef model. (b) Coarse registration result by KSR based method. The different views are shown in different colors. (c) The coarse regis- tration result shown in one color for visual clarity. (d) The final 3D model in the UWA Dataset (Ground truth). Figure best viewed in color. Performance evaluation of the proposed technique against popular spin image:
In Eq.12, mr is the mesh resolution of the 3D point-cloud. Table 1 reports the Table 1: 3D modeling accuracy achieved by the proposed KSR on Stanford dataset. A comparison with RoPs descriptor is also provided. 3D-modeling accuracy achieved by the proposed technique on Stanford 3D mod-
In Eq.12, mr is the mesh resolution of the 3D point-cloud. Table 1 reports the Table 1: 3D modeling accuracy achieved by the proposed KSR on Stanford dataset. A comparison with RoPs descriptor is also provided. 3D-modeling accuracy achieved by the proposed technique on Stanford 3D mod-
smaller compared with the average mesh resolution, the modeling process car Table 2: Modeling errors (in mr) of the four objects of the UWA dataset.
smaller compared with the average mesh resolution, the modeling process car Table 2: Modeling errors (in mr) of the four objects of the UWA dataset.
Figure 10: Achieved recognition results on three sample scenes from the UWA datsaset. The correctly recognized objects have been superimposed with their complete 3D models from the object database.
Figure 10: Achieved recognition results on three sample scenes from the UWA datsaset. The correctly recognized objects have been superimposed with their complete 3D models from the object database.
Table 3: Precision and recall values on the Ca’Foscari Venezia dataset. 7.2. Recognition Results on Ca’Foscari Dataset
Table 3: Precision and recall values on the Ca’Foscari Venezia dataset. 7.2. Recognition Results on Ca’Foscari Dataset
improve many folds once the algorithm is implemented in C++, which will make it Table 4: Computational efficiency of the proposed technique in comparison with state-of- the-art approaches.
improve many folds once the algorithm is implemented in C++, which will make it Table 4: Computational efficiency of the proposed technique in comparison with state-of- the-art approaches.
Journal of Computer Vision (2013) 1-17. selection process for 3d object recognition in cluttered scenes, International
Journal of Computer Vision (2013) 1-17. selection process for 3d object recognition in cluttered scenes, International
Related topics:
MathematicsComputer ScienceArtificial IntelligenceComputer VisionElectrical and Electronic EngineeringRepresentation Politics
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved