When we talk about an autonomous vehicle or any robot navigating its environment, perception fundamentally is about understanding the 3D space around it. For a self-driving car, this boils down to identifying what areas are drivable or navigable, and which are not. We need to build an instantaneous 3D understanding of the world relative to our own vehicle – the “ego” vehicle. Automotive computer vision stacks typically use machine learning (ML) based techniques to identify objects of interest, understand behavioral intent, and track dynamic objects in the scene.

I sometimes feel lucky to have experienced this surge in technology while I grew up – radio, cassettes, winamp, CDs, and now streaming. Looking back, radio was always fun to listen to, with some adverts peppered in between songs. After I had purchased my first car in 2017, I decided to give the radio a listen again. The advertisements were still there, and I found myself skipping through stations for most of the drive just to hear music. After 15 years, it felt as if nothing had changed, except that there were a lot more ads. What if I could have an program that automatically tunes to a station playing music?

Usually, when one thinks of encryption, they think of data being locked within a safe. You can transport this safe around but can only access it if you have the key. Information is secure this way. During my second summer during my graduate degree, I was introduced to Homomorphic Encryption – A new encryption scheme that not only kept information secure but allowed mathematical operations on data encrypted using HE. Information can not only be kept secure but transformed as well. This opens so many exciting avenues for the world of secure computing. If one wants to process sensitive data, a general concept would be to create a sandboxed environment with some network security. One could even set up an entirely offline environment. Regardless, we have seen frequent data leaks and thefts.

Introduction

This project experiments with scanning real-life objects and generating its 3D model in virtual reality. Using a depth sensor, and circling the object of interest, we can generate a point cloud for a set of angles while the corresponding texture/color map. After some cleaning, the individual point clouds are stitched to create a 3D model. The last step would be to render a mesh over the obtained points. As a constraint, this project may be limited to scanning objects with simple geometrical shapes and using sensors at fixed locations. I did this project as a part of my Computational Geometery course taught by Dr. Renold Bailey.