Research On 3D Vision Of Robot Night Vision Based On Infrared Video

In the field of robotics,a lot of artificially difficult operations in harsh environments are usually carried out under no light at night.The robot’s night vision field and scene recognition ability are directly related to the robot’s ability.In the night mode,infrared imaging system is usually used to obtain night scene images,infrared images with no color,low signal to noise ratio,low contrast and lack of depth,and so on.From the night infrared video to restore the scene and the object of the three-dimensional model is a field of computer vision research,it requires a comprehensive computer science,signal processing and other scientific knowledge,and it has a very broad application prospect in the visual navigation,military,industrial and other fields.Therefore,it is of great significance to carry out 3D reconstruction of scene at night.The main contents of the thesis are divided into three parts: The first part is the overview and the background of the present research and the basic theory of the three-dimensional reconstruction algorithm.The second part mainly introduces the commonly used camera calibration algorithm and infrared video preprocessing algorithms as well as direct sparse odometer.The third part uses three-dimensional reconstruction algorithm based on direct sparse odometer to reconstruct the scene of infrared video.The innovation of this topic is the first time to achieve the three-dimensional reconstruction for a single-sided infrared video scene.The reconstruction algorithm uses direct sparse odometer.In this method,the internal parameters of the thermal imager are obtained by the calibration of the infrared camera,and then the direct and sparse visual odometer model is constructed.Visual odometer front-end performs frame management and point management tasks.Using the sliding window and optimizing the total photometric error with the Gauss-Newton iteration calculate all the variables that the direct and sparse visual odometer model rely on,completing the task of locating the thermal imager and plotting.Experiments show that the method can achieve the three-dimensional reconstruction of monocular infrared video in real time accurately.