Autonomous Target Recognition And Grasping Of Industrial Robot Based On Multi-area Local Adaptive Descriptor

Vision-based intelligent industrial robots can perceive the external environment,quickly respond to changes in product model and size,making flexible production possible.Recognition and grasping of randomly placed objects are the most common tasks of intelligent industrial robots.It is a fast and effective method to recognize and grasp target by extracting discriminability descriptor from the image.The discriminability of descriptor is the bottleneck for the descriptor based method.The binary descriptor has been widely used,because of compact storage and the fast speed in extraction and matching.However,the low information density stored in each bit and the same sampling pattern for the all descriptor has limited the discriminability and robustness of the traditional binary descriptor.Therefore,in order to solve this problem,this paper proposes an industrial autonomous recognition and grasping method based on multi-area local adaptive descriptor(MALAD).The first chapter introduces the key technologies related to the target recognition and grasping with vision-mounted industrial robots.This chapter summarizes the status quo of researches on vision based target recognition and grasping,image enhancement,keypoint descriptor and keypoint based target recognition.Then we introduce the main research content,and the organization structure of this thesis.The second chapter introduces the adaptive images enhancement.Firstly,the image is decomposed by the first-order wavelet transform.Then,the low-frequency subband is transformed by using the Singular Value Decomposition and all the sub-bands are transformed by using the adaptive threshold function with the parameters optimized by using the Beatle Antennae Search Algorithm.Finally,the enhanced image is reconstructed by using the inverse wavelet transform.The third chapter introduces the extraction and matching of the multi-area local adaptive descriptor.The extraction of the multi-area local adaptive descriptor is mainly composed of two steps:the global-based sampling pattern construction based on scores of binary tests and local adaptive sampling pattern construction based on voting.We compared our result with BRIEF descriptor,ORB descriptor,BRISK descriptor and FREAK descriptor,which verified the high discriminability and robustness of multi-area local adaptation descriptor.The fourth chapter introduces target recognition and grasping by using the mean shift algorithm and the convexity and concavity consistency principle based random sampling consensus algorithm.By clustering feature points using mean shift algorithm,the number of iterations to get the correct model can be reduced.By using the improved random sampling consensus algorithm,the object with the maximum matching pairs in the template library will be considered as the recognition result.Finally,we get the accurate positioning of the target by using the contours of the target,the hand eye matrix and robot inverse kinematics.The fifth chapter develops an industrial recognition and grasping system based on multi-area local adaptive descriptor.The hardware components,software structure and the interactions between the modules are introduced.Then,we integrated the system using the software development kit provided by the robot and camera manufacturers.Finally,we verify the effectiveness of our system by a recognition and grasping experiment.The sixth chapter summarizes the full text and discusses the areas need to be further improved.