Research On Laboratory Virtual Intelligent Housekeeper System And Key Technologies Of Object Detection

With the development of disciplines such as computer science,information theory,cybernetics,and neural networks,artificial intelligence technology that began in the 1940s has made breakthroughs in recent years,including image and speech recognition,natural language processing,and machine translation based on deep learning Began to enter practical,which formed a wave of industrial revolution of "artificial intelligence+",quickly penetrated into various industries.Humanoid robots with higher intelligence are one of the ultimate goals of artificial intelligence.Although many research institutions in the world are engaged in research and development in this area,it still takes a long time to get into use.The laboratory tried to replace the humanoid robot with a six-axis trolley,focusing on information collection,communication,and processing to create an open and lightweight intelligent housekeeper system.This paper focuses on computer vision-based target detection and virtual reality-based VR display and interactive management.The main research work and results are as follows.The thesis developed a laboratory intelligent housekeeper software system based on a light-weight smart car,including a developer debug version based on MFC,OpenCV and MySQL,and a VR version intelligent housekeeper software system based on Unity3D and Steam VR.VIVE's VR hardware system.Aiming at the characteristics and needs of laboratory item management,the paper proposes a fast object detection algorithm based on image feature matching.The method first selects the ORB algorithm with high real-time and accuracy for image feature point extraction;then removes the mismatched points by Lowe's algorithm,and then filters the true matching points by Ransac algorithm;The cumbersome contour comparison measurement enables rapid detection of pre-registered items.The paper conducted a lot of experiments on the applicable conditions and accuracy of the algorithm,and on this basis,put forward the article registration and storage specifications.The thesis deeply analyzes and compares several target detection algorithms based on deep learning,selects the YOLOv3 algorithm as the target detection algorithm model in the intelligent housekeeper system,uses the VOC 2007 data set as a template for the training data set,and uses the Tensorflow framework for model training And adjust the parameters according to the actual situation of this paper to optimize the model.Experiments show that the target recognition algorithm based on YOLOv3 implemented in the paper has good generalization characteristics and can recognize people and trained items appearing in the video screen in real time.This method can be used as a kind of leading detection,that is,to first identify the type of item,and then use the above-mentioned fast algorithm based on image feature matching to identify individual items,greatly reducing the range of items required for the latter comparison detection.With the help of hardware equipment such as ZED camera,SENSE 3D scanner and depth camera,the thesis explores a variety of 3D data acquisition and modeling techniques based on binocular stereo inspection,structured light and time of flight.The modeling of large-scale three-dimensional scenes such as virtual labs in a real laboratory environment provides a foundation for VR display and interactive management based on virtual reality.The experiment shows that the laboratory intelligent housekeeper system based on the light intelligent car developed in this paper works steadily,and the key functions such as item and person recognition meet the preset requirements in terms of detection rate and accuracy,which basically meets the needs of the laboratory intelligent housekeeper system.The system is an open system.With the addition of more modules,its functions will be more and more perfect,and the degree of intelligence is getting higher and higher.