In this paper, we present a computer vision system for image
statistical modeling, used to achieve flight autonomy of a Micro Air
Vehicle (MAV) - a small aircraft equipped with a camera. The
autonomous control and stabilization of a MAV is based on real-time
horizon tracking, using the on-board vision system. Our objective is
to statistically model sky and ground and to incorporate the model in
the horizon tracking algorithm. The enormous variety of sky and
ground patterns renders color information insufficient for accurate
modeling, making us resort to texture analysis tools. Thus, we
choose, beside hue and intensity of the HSI color space, the complex
wavelet transform (CWT) for the feature space of our statistical
model. The Hidden Markov Tree (HMT) model is particularly well suited
for the CWT's inherent tree structure, as already intoduced in
literature. Therefore, we implement the HMT model and obtain reliable
and robust image segmentation, which we demonstrate on MAV flight
images. Our statistical image modeling may contribute to autonomous
performance of any intelligent system equipped with a camera. Its wide
range of applicability is illustrated also for an intelligent,
autonomous lawn mower.
|
