A robust adaptive horizon recognizing algorithm based on projection

Abstract

The main contribution of this paper arises from developing a modified adaptive horizon recognizing algorithm. The existing projection algorithm can only deal with those horizons whose projections are the global extrema. Based on the analysis of the projection peaks, new information was found to determine the horizon and three important indicators were extracted. An adaptive algorithm was proposed to recognize the horizon by utilizing these indicators properly. With the new criteria, more than 95% of the original algorithm failures can be overcome, and they are more robust when dealing with noise-polluted images. A speeded-up method was also derived. Verified with real MAV videos, this global searching algorithm satisfies real-time requirements, and the horizon recognizing correctness ratio is more than 99.9%.