Research On Image Matching And Location Technology Based On Binocular Vision

With the development of machine vision, binocular vision has been widely used in workpiece recognition and location, robot obstacle avoidance and navigation, three dimensional ranging, etc. Compared to monocular vision, binocular vision can acquire 3D information of the object more flexibly with more obvious advantages, and the operation is simple and easy to realize. Most research about the three-dimensional location of workpieces are based on the matching of SIFT, SURF, Harris and other feature points, where the detected feature points are few resulting in difficulty to ensure the location accuracy. Considering that workpieces are lack of texture and color features, binocular camera calibration and stereo matching technology based on edge feature are studied in this paper, in order to realize the 3D reconstruction of key grasping points and the orientation of workpieces.Because traditional nonlinear optimization algorithm is difficult to achieve stability and high precision in stereo calibration, a new stereo calibration method based on Hough transform and chaotic particle swarm is proposed. The Hough line fitting is used to modify the sub-pixel coordinates of the dot centers by ellipse fitting, improving the extraction accuracy of dot centers. During the nonlinear optimization period the particles are updated using the local optimal fitness value instead of the global optimal fitness value, and the corresponding local optimal positions within the neighborhood of particles are optimized by chaos particle swarm, as well as introducing in a globally adaptive dynamic inertia weight, solving the problem that the original PSO algorithm is easy to fall into local optima, which greatly improves the accuracy of stereo calibration.Because the mismatch due to the gray similar area cannot guarantee the location accuracy, a stereo matching method of workpieces based on improved shape context is proposed. Candidate matching points set are obtained using the histogram distribution of shape context, greatly reducing the calculation complexity; To increase the discrimination between the matching points and non-matching points, the similarity measurement of shape context is weighted by a weighted coefficient; During the fine matching period, the improved shape context is combined with the histograms of oriented gradients of the corresponding contour feature points in 3*3 neighborhood to acquire the initial matching point set; RANSAC is adopted to eliminate false matching points. The experimental results show that the proposed method is of high accuracy in matching and disparity computation.During the 3D location of workpieces, the 3D reconstruction algorithm based on parallel configuration of the binocular vision model is adopted, including three-dimensional reconstruction of the key grasping points, as well as the determination of the corresponding orientation through the angle between the principal axis of workpieces and the horizontal image axis. Besides, experiments of different workpieces are conducted, the test results show that the realization of the whole process is simple, and can satisfy the demand of the accuracy of subsequent grasping operation, which has certain practical value in engineering.