Caterina Massidda

I am researcher in the the  group.

My background is in Environmental Engineering and I am currently working on the application of multi-spectral analysis techniques to aerial robotics and multi-robot systems.

Please refer to the for details on my research topics

Robotics application of online multi-spectral analysis (May 2014-present)

Introduction

Multispectral analysis is a powerful and well known technique to classify material, surfaces, vegetation and soil. It is possible to use it remotely or in direct contact with the studied object. On aerial vehicles, its main application is environmental monitoring, hence is unrelated with the actual functioning of the robot. However, its employment in the decision process of a robot can greatly reduce the application of computationally-hard computer vision algorithms.

Goals

The goal of this project is to apply online spectral analysis to detect materials and surfaces. This will be done in two scenarios:

(1) from a short distance in swarm systems to identify objects and the target of a common action

(2) from a mid and long distance on autonomous Micro Aerial Vehicles (MAVs) to improve their understanding of the ground (e.g.: for online identification of safe landing areas).

However, these kind of sensors can be too expensive and too heavy for their equipment on MAVs. A viable solution, is the in-house development of low cost sensors, as proposed e.g. in [1].

Methods

A Near-InfraRed (NIR) single-pixel single-wavelength relative reflectance sensor can be easily integrated with a microcontroller. It can be used in scenario (1) to collect short range reflectance data of several objects, to be used for successive recognition in robotic swarms.

A general purpose low-cost low-weight multi-spectral camera can be implemented for scenario (2) either modifying a normal camera or employing an array of low cost CMOS camera modules in combination with appropriate filters. The definition of appropriate spectral indexes (i.e.: by combining the measured reflectance at different wavelengths) will highlight with low-computational effort the different types of ground.

Initial results

For scenario (1), the initial results, performed on values known in litterature and described in [2], prove that spectral measurements can be useful for recognition of objects in swarm systems.

For scenario (2), a prototype of a 2x2 camera array have been implemented. Inital sensor measurements shows promising results (see Figure).

References

1.\tDowning J, Murray AA, Harvey AR, low-cost multi-spectral imaging camera array, Int. Conf on Computational Optical Sensing and Imaging, Monterey, CA, USA, June 2012.

2. Stegagno P, Massidda C, Bülthoff HH, object recognition in swarm systems: preliminary results, ICRA 2014 Workshop: On the Centrality of Decentralization in Multi-robot Systems: Holy Grail or False Idol?, Hong Kong, China, June 2014.