Investigation And Implementation Of Electronic Image Stabilization Technology For Infrared Image Processing On On-board Platforms

With the development of maritime transport,fisheries and exploitation of marine resource,the frequency of marine accidents has also increased.The harsh maritime environment leads to difficulties in rescue.In China,there are deficiencies in deep-sea rescue,especially in the search technology for targets in distress at sea when a marine accident occurs.Therefore,it is important to develop a stable,fast and accurate On-board Maritime Distress Target Search System.Infrared camera imaging is the technology of choice for maritime search and rescue due to its ability to work around the clock without the constraints of harsh sea conditions.The combination of infrared cameras and rescue helicopters can increase the success rate of target searching in distress at sea.The bumps and shakes generated during the helicopter's rescue mission might cause the sequence of the collected infrared images to jitter,affecting the human eye's identification of the target in distress,and it is not conducive to the subsequent detection,positioning and tracking of the.target against the infrared images.Therefore,the use of fast and efficient image stabilization algorithms to correct jitter in the infrared image sequences generated by the On-board Maritime Distress Target Search System is of great practical significance.This paper investigates the key technologies for electronic image stabilization,including motion estimation algorithms,motion filtering algorithms,and motion compensation algorithms.The accuracy of the motion estimation algorithm is one of the important factors in image stabilization.In order to ensure image stabilization,the motion estimation algorithm must be made to calculate the offset between two frames as accurately as possible.In this paper,popular motion estimation algorithms,including grayscale projection algorithms,light flow algorithms,and feature point matching algorithms,are analyzed.In combination with the characteristics of the on-board infrared images,the motion estimation algorithm based on the Harris corner point was finally chosen.However,using this algorithm has the problem of not detecting a sufficient number of evenly distributed corner points for infrared images with poor imaging quality.To solve this problem,this paper proposes a method to detect corner points using an improved Harris corner response function combined with distance constraints,and further position coordinate correction is performed on the detected corner points,followed by trace matching using the proposed keyframe reference method combined with a multi-scale pyramidal light flow algorithm to complete motion estimation.The motion filtering algorithm removes the high frequency jitter component obtained in the motion estimate,and retains the active motion component of the camera.In this paper,a motion filtering algorithm based on Kalman filter is used to obtain the active motion component.The motion compensation algorithm calculates the motion compensation vector against the active motion component derived from the motion filtering algorithm,and then shifts the image based on the affine transformation model to compensate the image and output a stable image sequence.Undefined pixel areas appear at the edges of the image after the image is shifted,giving the image a black edge and affecting video 'quality.In order to solve the problem that the traditional cropping method reduces image resolution and the image splicing method is difficult to achieve real-time processing,this paper proposes a neighborhood-based compensation algorithm for undefined areas.By taking the rectangle with the largest area in the image display area and treating the area outside the rectangular image display area as an undefined area,the 1/n neighboring image area of the undefined area width is compensated,retaining the resolution of the target area and solving the problem of real-time image stabilization.The experimental results show that the method used in this paper is capable of real-time processing of infrared image sequences with a frame rate of 50fps,and the motion estimation accuracy can reach sub-pixel level.After the image sequence removes jitter,the human eye is more comfortable when viewing the target in the image sequence,and the stable image sequence provides assurance for subsequent infrared image target detection,positioning and tracking.