Research On Real-Time Mosaic Technologies Of Frame Thermal Infrared Images

Infrared airborne remote-sensing imaging is able to carry out all-weather, quasi real-time and three-dimensional observation of land, ocean and atmospheric targets, and has a great significance of both obtaining the information of ground floor resources and protecting national security. Infrared airbone remote-sensing imaging has developed for over 50 years. With the development of infrared material and crafts, infrared detector has undergone the stage of unit detector、linear array and plane array, at the same time, the imaging method of infrared airborne remote-sensing has experienced the period from scanning of unit detector、push-broom of linear array、multiple scanning of linear array to frame-sweep of plane array. Airborne infrared remote-sensing imaging is now developed toward the direction of high spatial resolution and high spectral resolution, along with the development of infrared plane array detector, obtaining a large field of view and high spatial resolution infrared remote-sensing images with use of infrared plane array detector has become a hot technology.Obtaining real-time thermal infrared images of large field of view to track and observe the target is of high application value in fields of battlefield reconnaissance and emergency rescue. The paper uses the remote-sensing images obtained by frame sweep imaging as data source and makes a thorough research of real-time image mosaic technology. The content and results of the research include: based on the deep analysis of the motion model of frame sweep imaging and aiming at the demand of real-time mosaic of frame sweep images, the paper proposes a scheme of fast image registration and mosaicing based on the point feature operator; through performance testing and validation of point feature operators, the paper selects a fast and stable point feature operator; using the relationship of image registration between sweep lines, the mosaic system recovers and optimizes the position and orientation parameters of images, and then completes the real-time refresh of mosaic image by registering and blending the sweep line to the mosaic image. In order to improve the real-time performance of mosaic system, the paper takes advantage of GPGPU to accelerate the algorithms and accomplishes a real-time mosaicing at a high frame rate.The main innovations of this paper are:1. The motion model of frame sweep imaging is analysed, and the number of image exterior orientation elements is simplified according to the motion characteristics. Based on the motion model the recovery scheme of exterior orientation parameters is designed and implemented based on the image registration relationship.2. A fast and stable point feature operator is chosen for image mosaic of frame sweep, and then the speed of matching image features is accelerated with the use of location sensitive hashing algorithm. The speed of image registration is improved under the premise of guaranteed matching accuracy.3. A simplified strategy of bundle adjustment is proposed to reduce the impact of cumulative error caused by image misregistration to the mosaic image, and to improve the recovery accuracy of image position and orientation parameters. Through fast and stable registration and blending of sweep lines, the real-time refresh of mosaic image is accomplished.4. According to the characteristics of each algorithm module in mosaicing system, the CPU+GPGPU platform is used to implement the algorithm in parallel. By division and cooperation of CPU and GPGPU, and the acceleration and optimization of algorithm suitable for parallel implementation by GPGPU hardware, the program flow is optimized and the real-time performance is improved.Finally, some proposals are made to the further research and engineering applications in real-time mosaicing of frame sweep images.