This project took place under the Institute of Automatic Control Engineering at Technical University Munich. idea of this project was to take a 'retina sensor', a new type of camera technology based on neuroscience research, and develop tracking algorithms for it. The retina sensor camera is an asynchronous visual sensor that instead of sensing light intensity values per pixel, senses if there was a significant change in intensity per pixel and asynchronously reports an 'event' for that pixel at the exact time of the change. This is very well suited for optical flow and tracking applications due to a faster response time and reduction in data. More info on this sensor can be found here.

This poses a number of challenges for tracking as there is only data available when something moves. A few different line tracking algorithms inspired from computer vision techniques were implemented and tested for this sensor. They are as follows:

• Hough transform using cartesian coordinates
• Hough transform using polar coordinates
• Random Sample Consensus (RANSAC)

The performance of all three methods were evaluated in terms of speed and quality. It was found that RANSAC was the fastest and most reliable.

The intended aim was to utilize this tracking technology to aid interactive segmentation techniques. Therefore multiple line tracking was also applied to track objects.

The following video demonstrates all the techniques. In this case the camera is fixed and a pencil is moved in front of the lense. The curvature is due to the lense distortion.

Presentation Slides