MyAV·¶

For safety, reliability and accuracy, robots must monitor the environment around them. They must be able to estimate their location and identify hazards and other disrupting influences.

This module will introduce the main concepts of how robots can process streams of data to discover geometric information about the environment and the objects within it. In particular, the module will focus on two of the most widely used and complicated streams of sensor data: images from cameras and depth sensors.

There are several ethical considerations to keep in mind regarding processing data in this context. By doing so, we can ensure that these technologies are used in a way that benefits society. This module will also support student in reflecting on ethical implications of data processing in computer vision and sensing applications and take steps to address any potential ethical concerns.

Aims:

The aims of this module are to:

• Develop students’ knowledge of the types of inferences and the algorithms used when working with single and multiple camera systems.
• Develop students’ knowledge of the types of inferences and the algorithms used when working with depth cameras.
• Support students in the development of a breadth of knowledge and understanding in the fundamentals of regression, classification, density estimation, dimensionality reduction, and model selection with the goal of applying this to data-modelling problems.
• Support students in reflecting on ethical implications of data processing in computer vision and sensing applications and take steps to address any potential ethical concerns.

Intended learning outcomes:

On successful completion of the module, a student will be able to:

1. Demonstrate a breadth and depth of knowledge and understanding of how data from raw camera streams can be used to create inferences about the geometry of the environment within which the robot is operating.
2. Select an appropriate control approach from a range of possibilities based on intended engineering objectives in the problem domain.
3. Design, decompose, plan and implement the algorithms for a range of suitable problems.
4. Identify the ethical implications of data processing in computer vision and sensing applications and take steps to address any potential ethical concerns.

Indicative content:

The following are indicative of the topics the module will typically cover:

• Structure of images and point clouds.
• Sensing technologies.
• Projective geometry.
• Monocular and stereoscopic imaging.
• Object primitive detection and fitting using RANSAC.
• Low level feature point matching.
• ML based approaches for feature point matching (SuperPoint, GCN, etc.)
• ML based approaches for other types of low-level analysis of point clouds and other sensing modalities.
• Pose estimation.
• Privacy, bias, transparency, consent and safety.

Requisite conditions:

To be eligible to select this module as optional or elective, a student must be registered on a programme and year of study for which it is formally available.