Visual Collision Detection for Flying  Vehicles

 

Project By:

Pedro Maia - pedroluz [at] seas

Oscar Nuñez - oscf [at] seas


Advisor:

Dr. Camillo Taylor - cjtaylor [at] seas

Jeffrey Byrne - jebyrne [at] seas

 

One of the major obstacles in developing small autonomous flying vehicles is the low power and weight constraints, because proven obstacle and collision detection techniques, like laser rangers and thermal imaging, requires heavy and power hungry hardware.


This project seeks to develop simple, robust vision based obstacle avoidance schemes that are amenable to the low power and low profile requirements of the vehicle. Spatial-temporal derivatives are used to compute time to collision estimates for every pixel in the field of view and used to classify the state of the vehicle as to whether it is in collision trajectory or not.


The correctness of the proposed approach was proved through unit tests and ROC curve performance analysis. The true positive rate detection of the system is currently about 93%. The functionality of the simulated  system implementation will be demonstrated through closed loop simulations.

Abstract