Study And Application Of Object Detection Algorithm Based On Template Matching And Neural Network

Object detection technology in digital images is widely used in many fields such as robot system,image retrieval and unmanned vehicle.In some practical industrial applications,the workpiece to be detected whose shape is single and data volume is scarce doesn’t have enough surface texture informations,which causes a great challenge to the accurate object detection.Therefore,it is of great significance to theoretical research and engineering applications how to predict object location,angle and other information accurately and quickly with using a small number of object samples in industrial scene.The main works of this thesis include three parts: object candidate box extraction method,object candidate box classification method,object detection algorithm application in the visual servo system.Through studying and analyzing the current object detection algorithms,this thesis proposes a object detection algorithm based on template matching and neural network.The algorithm in this thesis is divided into two stages,the first stage algorithmin generates candidate box quickly base on region orientation compressed.This stage algorithmin is an improvement on the template matching algorithm OCM,it modifies the compressing orientation method and measuring similarity method which are the core parts of the algorithmin.Compared with OCM,the improved algorithm can reduce 6.5% missed detections when generates approximate quantity of candidate boxes.The second stage algorithmin takes the region similarity scores between the candidate box and different model templates as input of the neural network.This stage algorithmin classifies and select candidate boxes base on neural network,which effectively reduces the false detections in the first stage.Through comparing the performance with other object detection algorithms,it is proved that the algorithm proposed in this thesis achieves a good balance between detection accuracy and detection speed.Finally,this thesis applies the object detection algorithm to the prototype system of container intelligent transport based on visual servo.According to the realization of hardware and software in the system and actual operation requirements,this thesis combines the proposed algorithm with a keyhole extraction method which is based on look-up table and achieves the operation goal of grasping the container accurately.In addition,in order to meet the real-time visual detection requirements of the system,this thesis achieves GPU parallel computing of the algorithm base on CUDA.Through practical test,the system is proved to be reliable and efficient for automatic container transport.