Image Colorization And Three-dimensional Reconstruction For Vehicular Infraredassistant Driving System

Vehicular infrared assistant driving system is becoming popular in civilian areas. Researchers have been paying more and more attention on the vehicular infrared image processing. Since the vehicular infrared image is monochromatic, low contrast and low resolution, which is difficult to recognition for drivers. Therefore it is very meaningful to make the information contained in the infrared image to be displayed in a better way.Both infrared image colorization technology and three-dimensional reconstruction technology can enhance the visual effect of infrared image. But so far these two technologies are not developed deeply and are not combined together to make a better result.In this thesis, we focus on infrared image colorization and three-dimensional reconstruction algorithm for vehicular infrared image. We realize vehicular infrared image colorization and compared the results with other infrared images colorization algorithms’ results. We optimize a three-dimensional reconstruction algorithm for visible light image based on characteristics of infrared image. Finally we realize colorful three-dimensional reconstruction of infrared image successfully.The thesis is divided into three parts. The first part is the introduction and implementation of the framework about vehicular infrared assistant driving system, and an introduction of present theoretical approaches about colorization and three-dimensional reconstruction. The second part introduces two proposed infrared image colorization algorithms. One is based on random forest optimization and super-pixel segmentation. The other is based on the label transfer algorithm. The third part is the algorithm of infrared image three-dimensional reconstruction which is based on the panel parameters image reconstruction, and the colored infrared image is performed three-dimensional reconstruction by the algorithm.The main innovations of this thesis are as follows:1. The vehicle infrared image colorization algorithm based on super-pixel segmentation and histogram statistics is proposed, which has obvious advantages of precision and efficiency compared with other colorization algorithms.2. We propose label transfer based infrared image colorization algorithm, which can solve the issue of the limited number of object classification in the infrared image. The algorithm can detect the objects in infrared image except from the sky, the ground, trees and pedestrian.3. The three-dimensional reconstruction method of panel parameter Markov color image is used in color image successfully. This algorithm is optimized according to the characteristics of infrared image. The advantage of this three-dimensional reconstruction algorithm is few assumptions of images, high adaptability characteristics, which can be very good for the vehicular infrared image three-dimensional reconstruction.4. The three-dimensional reconstruction algorithm is combined with the infrared image colorization algorithm. The colored infrared image is put into the model of three-dimensional reconstruction to build the 3D scene. Through the above process, the effect of infrared image is more intuitive. The processed infrared image has not only colors, but also the stereo visual effect. The experimental results show that the 3D reconstruction of vehicle infrared image can achieve good results similar to the color daytime image.