How does a fly avoid obstacles?

Abstract: An implementation of a Time-To-Impact(TTI) algorithm in a language more closely related to small compact devices, such as a Field-programmable gate array(FPGA), needs tobe developed. The algorithm is based on an article written by Åström and Forchheimer [1].The new implementation in C++ is evaluated towards a previously developed Matlab implementation and the performance increase of the C++ implementation for this particularalgorithm is measured. The Time-To-Impact algorithm is evaluated in this thesis using aportrait of a person, a self produced image of a whiteboard, a sequence of a sled moving towards the camera and videos of a car approaching the camera. The first tests of the portraitand whiteboard are done using Matlab. After this a sequence is generated with the imagesof the portrait and the whiteboard. These are evaluated in a Matlab and C++ implementation. It is in this step the performance difference is measured. The sled sequence used is ina controlled environment and evaluated using the Time-To-Impact algorithm translated toC++ and adjusted to work on a sequence. Lastly the video of the car is evaluated using thesame C++ implementation from the sequence adjusted for a video. For the sled sequenceand the car videos a TTI is calculated and evaluated against a ground truth generated fromthe known distances and velocities. The results indicate that the algorithm works in controlled environments and for real life videos. The results also indicate that for some framerates the TTI is not correct in comparison to the ground truth, and that the algorithm cancorrect this with adjustments e.g., by using more filtering.