The Research And Application Of The Algorithm For Recognition Of Multiple Target With Rotation And Scale Invariance

The target recognition technology based on machine vision is an emerging technology based on computer vision and image processing. It extracts various characteristics of the target to be identified by image processing method, and then understands and matches the characteristics,so ultimately achieves final targets of automatic recognition and classification. It is also one of the important research techniques of realizing industrial automation production. Due to the problems of target's attitude angle and the height of camera changing, the target in the camera's view of vision will change in terms of angle and scale. For another hand, there will be a variety of targets in a mixed line of production, so the research of multiple target recognition technology with rotation and scale invariance has become one of research focus on the target recognition technology based on machine vision.In this dissertation, the goal is to improve the speed and accuracy of multiple target recognition with rotation and scale invariance. The target recognition algorithm based on machine vision is studied and improvement, and the mainframe of the dissertation is arranged as follows:First of all, the aim is to identify not only the target category and location, but also the target's attitude angle, using the target recognition's technology based on machine vision to solve specific industrial problems. And the principal axis of target will be extracted based on the target's contour, so the rotation angle of principal axis can be taken as the target's attitude angle.Secondly, aiming at the deficiency of the existing target recognition algorithm based on contour corner in accuracy and anti-noise ability, the author proposes a multiple target recognition algorithm with rotation and scale invariance based on characteristic string of corner angle. The proposed algorithm constructs the characteristic string of corner angle and identifies target using the target contour corner and main spindle to. The experimental result shows that the proposed algorithm has certain advantages in terms of speed, accuracy and anti-noise performance.Thirdly, aiming at the deficiency of the insufficient in target's adaptability of contour recognition algorithm and the poor real-time performance of recognition algorithm based on the content of the target area, the author proposes a multiple target recognition algorithm with rotation and scale invariance based on target's contour and region segmentation algorithm. The proposed algorithm constructs the feature vector for target recognition by extracting regional feature points using equal interval line segmentation of target area. The experimental resultshows that the proposed algorithm has not only strong adaptability of target, but also great advantages in real-time performance.Finally, the proposed algorithms are applied to the target's automatic marking with high-speed pipeline in the industrial application. The research results also can be extended to other industrial field of on-line identification with a broad prospect in application.