Research On Multi-feature Fusion Template Matching Method Based On NNF

Template Matching(TM)algorithm is a classic method in the field of target detection.The key to the task is to match the given template in the image to locate the target.With the advent of the 5G digital age,the importance of fast and accurate target detection algorithms in production and life is increasing day by day.Computer vision replaces human vision for target search and positioning,showing strong advantages in national defense,autonomous driving,production and life,etc.The field has a wide range of application scenarios.Traditional template matching methods are mostly used in industrial scenes,unable to deal with increasingly complex scenes.Therefore,effective template matching algorithms for multiple difficult natural scenes need to be proposed urgently.This paper proposes a similarity measure template matching method(Nearest Neighbor Field-based Salient Feature Fusion,NSFF)based on nearest neighbor feature extraction and multi-feature fusion,which effectively characterizes the target of each part of the scene image by constructing and filtering multiple features and then meet the needs of local feature extraction for template matching methods through fine and robust feature measurement so that it can still describe the target robustly in deformation,rotation and severe lighting changes.This paper further integrates and measures multiple features through linear weight parameter fusion,and gives the center coordinates and calibration frame of the largest possible area.In this paper,the application scope of the proposed method is further expanded,so that it can adapt to the application scenario of single target and multiple templates and adaptively change the size of the bounding box.The features extracted by the algorithm in this paper fully describe the global pixel value and structural semantic information.The experimental results prove that the fused features can increase the target attention of the algorithm and effectively resist the interference of the background.In the experiment,the accurate target position and performance Robust results under multiple difficulties.The algorithm has a success rate of 70.5% on a complex data set with an image resolution of 480×360,and the detection accuracy Mio U reaches 0.572.