Research On Tracking Related To Surface Target Based On Improved SAMF Algorithm

Computer vision is a science that has risen and developed rapidly in recent years.And target tracking technology is a key technology in computer vision.Target tracking has important research significance and practical applications in military defense,industrial production,civil security and other aspects.Unmanned Surface Vehicle(USV)is a new marine equipment.It has great significance for maintaining marine sovereignty and exploiting marine resources.Applying target tracking technology to surface target tracking,in conjunction with the USV,will help the USV to perform tasks such as guard patrols,combating piracy,maritime search and rescue,obstacle avoidance navigation,and hydrogeographic surveys.We must consider the actual environment of the water surface when we apply target tracking to surface target tracking,Water surface targets usually have problems such as target blurring,jitter,and occlusion which will affect the target tracking effect.Based on the excellent performance of correlation filtering algorithms,this paper compares the tracking performance of different correlation filtering algorithms.The SAMF(Scale Adaptive Kernel Correlation Filter Tracker)algorithm has good adaptability to the problems of target rotation,deformation,illumination change,scale change,etc.Combined with some problems of surface targets,the original SAMF algorithm was improved to adapt to water surface target tracking.The main research work of this paper are as follows:(1)For surface targets,such as warships,speedboats,fishing boats and other targets,there is a problem of high-speed movement.The original target tracking algorithm can’t adapt to the fast movement of the target and lost the target.This paper improves the filter model updating mechanism of the algorithm,so that the template updating rate is no longer a fixed value,but can be adjusted according to the target speed adaptively and modify in advance according to the acceleration..The faster the target moves,the faster its characteristics of the appearance change.By increasing the update rate of the template and updating the change to the filter model,the tracking effect of the target can be maintained.Through a lot of experimental research,the relationship between template update rate and target speed and acceleration is obtained,and the self-adaptive adjustment rule of template update rate is determined.(2)Aiming at the problem of frequent wave occlusion or other obstructions in the water surface environment,this paper improves the model updating method of the original algorithm.In the process of correlation filtering algorithm,after tracking a certain frame of image,target features will be extracted to train the filter and update the filter model.When the target is occluded,the extracted target features have been polluted.If the filter model is continuously updated,the pollution will be introduced into the filter model.Occurrence of occlusions will continuously accumulate errors and eventually affect the tracking performance.Therefore,the update of the filter model is stopped during occlusion.By tracking the image sequence containing occlusion,the improved algorithm improves the recognition rate and tracking effect.(3)In order to meet the real-time requirements of the algorithm,this paper improves the target search area of the algorithm.Generally,when tracking,the search is performed within a certain range around the target in the previous frame.This range is generally 1.5 or 2 times the target area.The improved search area is adjusted according to the target size and the ratio of length and width.This step will make the search area more concentrated around the target,and at the same time can remove the unnecessary background area.By tracking the actual image sequence,the improved algorithm improves the tracking speed by 3 to 4 frames per second compared with the original algorithm.