Undergrad Research Project - Road surface classification on large data sets

Spring 2016

Student
Nandini Ramakrishnan
Advisor
Martial Hebert
Project description

The goal of this project is to classify the roughness of roads and find road objects like potholes, manholes, speed bumps, and expansion joints in roads. The initial algorithms have already been developed to do this for a limited dataset. The data consists of GPS, video, accelerometer and gyro data collected from a vehicle and the algorithms use the accelerometer and gyro data to do the classification. This will be expanded to work on large datasets with a large variety of roads, objects, vehicular speeds, etc. To do this I will make use of various algorithms found in the literature and algorithms developed myself. The tools to qualitatively and quantitatively evaluate these will also be developed. E.g. a script should find the images that correspond to the found object and it makes it possible to visually confirm the findings. An example of a quantitative measure is a precision-recall curve. The algorithm has to be efficient enough to analyze our collected data within a reasonable time.

Return to project list