Image Recognition And Measurement Based On Binocular Vision System

Human beings can obtain a large amount of information from outside through various senses.Among them,the most important thing is to obtain information through vision.The information that humans obtain through vision accounts for the vast majority of all information.The visual system is able to obtain information about the outside world through the eyes,which means that the human eye can convert two-dimensional information into three-dimensional information.This is very important for us.We can use this principle to obtain the three-dimensional information of the object with the camera.After obtaining the three-dimensional image information,how to effectively extract the information of the target object we need,it is necessary to introduce deep learning.The method enables the machine to automatically identify a specific target object through a convolutional neural network.This project aims to realize the goal of extracting the target from the complex background and removing the background through the network of deep learning after capturing the image through the camera,and then integrating the depth map and the color image on the target,and realize the length of the object after clicking any two points on the image.In the animal husbandry,the measurement of the length or width of the object is often carried out,and if the direct measurement method is used to cause the excited reaction,the method can realize the indirect measurement of the object by the image acquired by the camera.First,the overall structure of the system is introduced.The principle of camera object imaging and the internal and external parameters of the camera are briefly introduced.The Intel D435 depth camera and the Intel D415 camera were compared.And introduced several depth map acquisition methods such as: To F method,structured light method,active stereo binocular vision method.Several convolutional neural network training models are compared.Secondly,the method of fusion of depth image and RGB image is introduced,and the depth map and RGB image are combined by Intel D415 depth camera,and the program of measuring the distance between any two points on the image is realized.We use SSD convolutional neural networks combined with migration learning and data expansion to train with few target training sets.,and the network model is embedded in the D415 system,so that the system can automatically measure the data such as the length and width of the object after the object is recognized.Finally,the system is applied to the livestock industry for measuring the body length and weight of pigs.