Project P9:
Finding Objects in 3d Point Clouds of Urban Environments

Project Goal

Stanford is working on digitizing a major city, such as San Francisco, using a truck with multiple lasers and cameras for data acquisition. The goal of this project is to analyze point clouds and images for the presence of things such as cars, buildings, people, and other objects of relevance.

Project Scope

Your software should accept as an input a point cloud and an aligned set of camera images, and output labels for each point. A good starting point is a recent paper by Taskar et al on training Markov Random field for object classification.

Tasks

• Software for parsing and displaying point clouds and camera images
• Software for running a MRF algorithm over the data
• A software system for manually labeling the data
• Implementation of the Taskar et al learning algorithm
• Systematic evaluation and display of the results.

Project Status

Julian Snyder (jboolean at stanford dot edut),
Eddy Hartanto (hartanto at stanford dot edu) (preliminarily),
Juan Sepulveda (sblvd at stanford dot edu).

Point of Contact

Sebastian Thrun

Midterm Report

not yet submited

Final Report

not yet submitted