Monocular Vision Research Of Pose Estimation Based On Circlar Features

Vision measurement is a widely used technology which involves many disciplines, like computer measurement theory, digital signal processing, image processing, pattern recognition, and computer technology, etc. With the development and application of computers in various fields, vision measurement technology has made rapid progress in relevant areas. Especially the pose measurement technology based on monocular vision is widely applied in many practical scenarios for its simple structure, stable performance, low cost and fast estimated speed.This paper is to achieve the task of measuring the positions and orientations of characteristic circle based on vision measurement, which are in a complex environment especially with haze. Firstly, theory of vision measurement model is introduced, as well as the internal and external camera calibration methods, which are commonly used, and selecting Zhang Zhengyou calibration method to calibrate the camera. Secondly, the circular target pose measurement algorithm based on monocular vision feature is discussed, the algorithm can calculate three-dimensional pose from two-dimensional images. But the result is ambiguous, in order to eliminate ambiguity, pose measurement algorithm based on sequential images is proposed. In real-time surveillance video, the PAL system(25 frames per second) is used generally. The pose of low-speed object has little change. Therefore the relevance of pose between successive images can be used to remove a false solution. Then, the appropriate test equipments and test layout is selected. Under the equipments, the images are pre-processed, including wavelet denoising, k-means clustering algorithm and elliptic recognition. The results demonstrate the effectiveness of the algorithm and meet the requirements. Finally, In view of the environment of the fog, dark channel prior is discussed, which can make the image more natual. But because of heavy fog, the recovered image has a certain degree of distortion, and can not be segmentated after defogging. To highlight the contrast of foreground and background, the improved dark channel prior algorithm is presented, which is more suitable for real-time target detection.