Title: CNN-based Single Image Obstacle Avoidance on a Quadrotor

Speaker: Dr. Prunarjay Chakravarty

Time: Tuesday, 27 Sep, 5pm
Venue: RM101

Abstract: This talk will be about the use of a single forward facing camera 
for obstacle avoidance on a quadrotor. A ConvolutionalNeural Network(CNN)
is trained for estimating depth from a single image. The depth map is then fed 
to a behaviour arbitration based control algorithm that steers the quadrotor 
away from obstacles. Experiments conducted demonstrate the use of single image 
depth for controlling the quadrotor in both simulated and real environments. 


Speaker's Bio: Dr. Chakravarty completed his PhD from Monash University, Australia 
in 2010, where he worked on using mobile robots for surveillance applications. He 
subsequently worked for Sensen Networks, a start-up where he led a team that developed 
a mobile parking enforcement system - a car equipped with a computer vision system 
that automatically geo-tags number plates of vehicles that are parked illegally or 
have overstayed their allotted parking. At Sensen, he was also involved in a video 
surveillance project at Abu Dhabi airport, which involved the development and setting 
up of the largest video analytics system in the world at the time - face and number 
plate recognition on feeds from 1800 cameras at the airport. After 4 years at the 
startup, he moved to KU Leuven, Belgium, where he is currently employed as a post-doctoral 
researcher. He works on several computer vision projects including video diarization 
and using vision for automating the flight of quadrotor helicopters. He has publications 
in both Robotics and Computer Vision conferences like ICRA, IROS, ICMI and ECCV.