Study On The Application Of The Multi-modal Image Registration Technology In The UAV Autonomous Navigation

Image navigation has been a primary trend in precise weapon guidance, including the visible light image navigation, infrared image navigation, SAR image navigation and other forms, navigation system integrated image navigation and other navigation methods posses higher accuracy and reliability. How to enhance the survivability of UAVs in real-time battlefield becomes a top priority issue of UAV navigation under the unfavorable factors of electronic warfare and anti-satellite warfare. Image navigation using infrared and visible light images as registration source not only has high autonomy and strong ability of anti-electromagnetic interference, but also has the advantage of night-time navigation. Therefore, UAV image navigation based on infrared and visible light image registration is an ideal and reliable navigation method.Because the differences exist between the infrared image and visible light images in characteristics, distance, time and point of view, the object reflected in the gray, shape and angle of rotation between images would be greatly changed, the change resulting in the image of local non-similarity, combined with the requirement of real-time of navigation, the traditional image-based gray and feature-based registration methods can not meet the requirement. Therefore, it becomes an urgent need to look for a kind of registration algorithm which is adaptable, highly precise, fast to UAV in a complex battlefield environment for autonomous navigation.This study focuses on applying registration algorithm as the key technology in navigation image based on the above issues. Image registration under complex cases were carried out using maximum mutual information and normalized mutual information on the bases of the characteristics of infrared and visible light images, taking mutual information as the similarity measure, effectiveness of mutual information measurement was confirmed as well in this study. A method to overcome the large-angle rotation is presented to solve the problem in wide-angle image rotation. The accuracy, robustness and near-real-time of the method were verify through simulation results of noise, blur and other cases.The nature of image registration needs multi-parameter optimization. Improvement of basic Powell was performed to overcome the shortcomings of optimization algorithm vulnerable to local minima and search inefficient. Because the interpolation, an important step in the image registration,would inevitably bring local extreme, the bilinear interpolation and the Partial Volume interpolation were combined to form the good complementation of two algorithms advantages based on the local maximum formation factors of mutual information function.In addition, on the base of the theory of wavelet, a detailed description of the wavelet registration with a number of issues is given. The hierarchical search strategy of image is carried out by improved Powell method and wavelet multi-resolution, this rapidly promoted the speed of the algorithm.