Based On Single Camera 360-degree Rotation Tracking System

Computer vision technology has been using more and more widely in the mobile robot, video surveillance and tracking, autonomous navigation, etc. Tracking the moving target-object is one of the key technologies in robot visual tracking. Except that it is required a system to be able to deal with images to identify the target location, but also required the system be able to depending on the target robot position to control exercise in order to achieve its goal of tracking.In this paper, the research work is based on the camera image setting, dealing with gray-scale image, selecting the image characteristics of the region and setting up the image coordinate system. In the continuous acquisition process, using image processing technology to image two image-matching, and correcting whether the image coordinates are consistent. If there are shift offset, system will send it to the processors, and then control the positive direction of rotation as well as motor speed-control according to the shift offset. So that the tracking target image will always be in focus as the camera monitoring center in order to tracking the object.In image processing context, this paper use color image of the pre-processing first, then use the SIFT feature matching method proposed by Lowe and others to carry out the two images before and after in the feature matching in order to realize the purpose of image tracking. This feature has excellences in rotating, zoom scale and brightness changes, etc. It also maintains a good uniqueness, informative invariance, and be able to provide the best matching feature points.It uses no-model tracking control in tracking control which can be arbitrary targets tracking. It can eliminate the steps of specific targets in modeling, then follow, which can increase the diversity tracking system greatly. It uses the Kalman filter algorithm in the algorithm, which is extracted from the signal-related concept of measurement through the algorithm to estimate the required signal a filtering algorithm. The advantage is not only a smooth one-dimensional estimate of the random process can also be of non-smooth, multi-dimensional stochastic process estimation.