Research On Object Detection Technology Of Robotic Vision Svstems

Object detection technology is one of the key technologies for robotic vision systems to perceive and locate objects in their visual field.It is the base for several senior vision tasks,for example,scene understanding.Researching on object detection algorithms in robotic vision systems and impoving them are significant for the improvement of the intelligent degree of robots.This thesis divides object detection technologies of robotic vision systems into three layers,which are rigid object detection,acquired object detection and general object detection.Then,research has been done on the three sub-topics according to the order of increasing difficulty and degree of generalization.For rigid object detection,this thesis presents two rigid object detection algorithms which can be used in two practical robots,heat-engine plant temperature detector assemble robot and Chinese chess game robot,respectively.The two algorithms have worked well in their corresponding vision systems.Afterwards,this thesis has done research on technologies used in acquired object detection,especially the application of convolutional neural network.This thesis presents an end-to-end convoluntional neural network model to detect second generation ID cards in natural images which has wide application prospect.In the task of detecting second generation ID cards in natural images,the model can achieve 79.51%average cover rate.In the end,the thesis has researched on general object detection and presents a general object detection algorithm based on segmentation-based structure feature.The algorithm yields 96.1%detection rate with 1000 proposals in the PASCAL VOC2007 dataset.Moreover,its detection performance outperforms 4 state-of-the-art methods when the number of proposals is less than 100.