Research On Multi-object Detection And Recognition Based On Microassembly System

With the advance of technology and high speed development in micro/nano science field, the miniaturization of device is going to be a trend in industrial manufacturing. At present, the micro-device is widely applied in lots of areas, such as the latest plan, “Breakthrough Starshot”, which put forward by Hawking has come up with the concept of micro spacecraft. The scale of working space under microscopic world is smal, and the size of operation object is measured in micro or even nanometer level. As a result, it is out the precision limitation using traditional method. Therefore, micromanipulator, as a kind of approach to research the micro world, has been paid more attention around the world. In this paper, we has designed a set of special robot system to assemble ICF fusion targets, and it is composed of micro vision which acquires the information about the location and posture of the objects and feeds it back. Meanwhile, the method of multi-object classification, recognition and location based on micro-vision is well researched.Firstly, the process standards and accuracy requirements of the target assembly task are analyzed, then the hardware architecture and software framework are discussed in detail in this paper. By the result of experiment, it turns out that target assembly task is well done using our robot system.According to the imaging characteristics of the microscopic image and the interference of the experimental environment, the error rate will be high when directly using the collected images to identify multiple targets in space. So it needs to use a series of preprocessing methods to weaken the influence of noise and highlight the characteristics of the targets. To begin with, graying arithmetic is used to reduce image data. Next, filter and morphology processing is used to eliminate noise, dust and filaments interference. Then, contour chains of targets are extracted by edge and improved contour detection algorithm. At last, segmentation algorithm is used to get outline of each target. Because of the prominent feature of object shape, the shape feature and the feature of affine invariant moments can be used to accurately represent the nature of the object in the system. In this paper, this combination features are selected as the basis for recognition.In order to avoid blind clamping and releasing to the target, the proposed algorithm of the object center location and pose estimation successful y calculated the center of a target and the pose angle, providing accurate data for the automatic assembly tasks. In the case of object occlusion, this paper proposes a recognition method based on ORB feature points matching. The local feature points are used to replace the global features of the target, which can achieve a good recognition effect.In multi-object recognition methods, the support vector machine has obvious advantages, which can be very good to deal with multi classification and nonlinear problems. In this paper, a multi object recognition system based on the combination feature of the target and SVM is designed. The experiment proves that this method has a better recognition result than other methods.