Download or read book Features for 3D Object Retrieval written by Cristina González Delgado and published by . This book was released on 2016 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: This project is focused on the object retrieval challenging field, a critical area for robotics and computer vision systems. Specifically, this project is trying to address some issues in the object recognition area, which is a key during the object retrieval process. Recently, low cost RGB-D sensors able to capture not only color information but also depth data, have emerged. This is a great opportunity to increment efficiency and robustness of object retrieval applications by taking advantage of this additional depth information. In this project I implemented an evaluation system of 3D feature descriptors by using the Point Cloud Library (PCL) , and used it to perform an specific assessment for two of the most widely used descriptors for object recognition: SHOTColor and PFHColor. They were tested in two different datasets containing data extracted with the Kinnect RGB-D sensor. The first dataset consist of 3D point clouds of individual objects and synthetically built scenes, the second one has 3D point clouds corresponding to individual model objects and scenes from the real world. The evaluation process was done considering the following three properties: descriptiveness, robustness to Gaussian noise impact and support radius variation, and efficiency. As a final result, the outcome of this evaluation process is the performance assessment on those descriptors depending on the different scenarios. SHOTColor descriptors performed better in terms of descriptiveness, robustness to Gaussian noise impact and support radius variations, and efficiency as well. But, they required higher storage capacity than PFHColor descriptors. SHOTColor descriptors seems to be more sensitive to clutter and occlussion effects, since the performance for the real scenes dataset is worsening faster than the PFHColor one. Given those considerations, SHOTColor descriptors are recommended as the best option for time-crucial applications and also for those ones requiring a strong descriptiveness power. On the other hand, PFHColor descriptors are better for spacial-crucial applications.