Fast target tracking method with improved SIFT algorithm

Abstract

The invention belongs to the field of target tracking, and in particular relates to a fast target tracking method with an improved SIFT algorithm. The fast target tracking method comprises the steps of: obtaining a dynamic video by a collection device, obtaining a target image from the dynamic video, extracting SIFT features of the target image and storing into a database, matching the SIFT features of the target with SIFT features of every frame image in the dynamic video, after detecting the target, determining the position of the target in the image, initializing a Mean Shift template according to the area of the target position, iteratively tracking every frame image behind the dynamic video by the Mean Shift, and determining the update frequency of the template according to the tracking target. The fast target tracking method achieves a very effective way for fast tracking objects, and can be widely applied to fields of video monitoring, movement analysis and the like. Compared with the traditional SIFT tracking method, the fast target tracking method with the improved SIFT algorithm is greatly promoted in speed and can completely meet the requirement of being real-time.

Description

technical field [0001] The invention belongs to the field of target tracking, and in particular relates to a fast target tracking method with an improved SIFT algorithm. Background technique [0002] Object tracking refers to finding the position of the moving object of interest in each image of a sequence of images. It is an important research direction in the field of computer vision and is often used in video surveillance, artificial intelligence, and human-computer interaction. Target tracking can provide the track of the monitored target, and also provide reliable data information for the target's motion analysis. Target video target tracking algorithms mainly include methods based on contrast analysis, methods based on feature matching, kernel methods, motion detection (optical flow method) and so on. [0003] Moving object tracking has broad application prospects in many fields such as intelligent monitoring, human-machine interface, virtual reality, motion analysis,...

AI Technical Summary

Problems solved by technology

In addition, the kernel function histogram is used to model, which is not sensitive to edge occlusion, target rotation, deformation and background motion, but the size of the window width remains unchanged during the tracking process, and the target template cannot be updated in real time. , the pose of the target, the light of the environment, etc. will change, and the candidate template of the target that has changed in the scene image is used to match the initial template, which will lead to tracking failure

David Lowe proposed the SIFT algorithm in 1999, and carried out further development and improvement in 2004. Since SIFT was proposed, it has attracted the attention of many scholars, because the SIFT feature matching algorithm can handle translation between two images, The matching problem in the case of rotation and affine transformation has a strong matching ability. With the continuous development of the SIFT algorithm, this method is currently widely used in workpiece recognition, medical image registration, mobile robot positioning and map creation, image Stitching, face recognition, 3D target retrieval and tracking, target recognition, texture recognition, wide baseline image matching and image feature matching, but the SIFT algorithm matching speed is slow and the real-time performance is poor

Images