object detection and 3d modeling · screened poisson surface reconstruction to create a 3d model....
TRANSCRIPT
-
Electrical and Computer Engineering
Object Detection and 3D Modeling
Mentor(s): Murali SubbaraoAnanth Rajan
3D Modeling of Indoor Scenes
-A 180-degree view of an indoor scene was captured and combined using the Iterative Closest Point algorithm toproduce a combined point-cloud.
-The combined point-cloud was sampledwith 30,000 points and reconstituted usingscreened Poisson surface reconstruction to create a 3D model.
Social Impact
-This technology has many possibleapplications such as cartography, lawenforcement, Augmented Reality, and videogame design. It has the possibility ofcreating a space for simulating realisticindoor conditions for educational/researchpurposes.
Background
-The Intel Realsense D435Camera has an integrated stereo depth and RGB sensor that allowsperspective to be added to an image frame.
-Realsense Viewerallows simultaneousRGB-D capture.
-Combined RBG and depth data isdisplayed In point-cloud format forprocessing.
-A point-cloud was sampled using PoissonDisk Sampling to reduce noise and improve clarity of generated 3D model.
Project Team Name: i3
2D Applications
-Face detection and featurerecognition was implemented using cascade object detection
-Object recognition using pretrained AlexNet neuralnetwork allows common objects to be classified.
-Panoramic stitchingusing feature-basedimage registration
Special thanks to Professor Murali Subbarao for guiding me through this project.
Glossary
Point-cloud- A set of points in 3D spaceRGB-D – Red, Green, Blue, Depth data