Development and Implementation of an Image-Processing-Based Horizon Sensor for Sounding Rockets

Abstract

In this thesis a new flexible and robust horizon sensor software for sounding rockets is presented. It operates on image data in the visible spectrum of light and provides two axis attitude information. After a short analysis of the current available technology, the novel algorithm is explained in detail and the new mathematical concept is thoroughly defined. It features a very specialised and optimized edge detection, outlier detection, fisheye distortion compensation and a plausibility check. Moreover, it can cope with strong disturbances introduced by the Sun and inhomogeneities of the reflectance of the Earth while maintaining a linear runtime in the number of pixels. No a priori knowledge apart from the altitude is required by the algorithm. With simulations using Google Earth Pro, the major influential factors to the accuracy and their impact are determined. The software in C++ was implemented for the DSP Blackfin 561 and the RODOS realtime kernel. A tool that simulates an ethernet camera was developed in order to validate the functionality of the software on the DSP. To determine the intrinsic parameters of an optical system, a program that analyses images of chessboard patterns was implemented using OpenCV. The evaluation determined an accuracy better than 1° in the simulation with a sample time between 0:2s and 1s, depending on the set of parameters. The minimum altitude of 25km and the rate of false positive detections of 5% were determined with real footage of a sounding rocket flight from GoPro cameras.