Research And Application Of Intelligent Robot Target Detection Technology

With the rapid development of intelligent manufacturing and the Internet of Things,more and more intelligent robots have shifted from their original factory environment to indoor working environment.However,their application is limited by the environmental perception ability of devices and still performs some traditional automation work.In order to address the current demand for richer environmental information for intelligent robots,this thesis proposes a fully autonomous target detection method that integrates multiple perceptual information,based on object detection algorithms,robot real-time localization and mapping(SLAM)algorithms,and depth camera spatial positioning algorithms.At the same time,in response to the problem of difficult recognition of target detection algorithms in indoor occluded scenes,this article chooses to improve the target detection algorithm and proposes a decoupled detector based on enhanced edge features.The main work of this article includes the following three parts:(1)In response to the difficulty of complete autonomy faced by current target detection tasks,this article combines SLAM algorithm,target detection algorithm,and depth camera coordinate system conversion method to integrate basic perception data such as RGB image perception information,depth perception information,and laser scanning information to achieve modeling of the search space.In addition to the core modeling algorithm,this thesis also proposes a circular filtering algorithm to solve the depth measurement problem problem? A corresponding threshold denoising method has been proposed to address the issue of a large number of duplicate results in detection results? Two brief metrics were proposed to quantify the performance of target detection algorithms.The algorithm proposed in this article has been proven to achieve fully autonomous indoor target detection tasks,but there is still room for improvement in the algorithm.(2)Based on the experimental results of(1),this study confirms the issue of insufficient accuracy of object detection models in indoor occlusion scenarios,which can seriously affect the final results of object detection tasks.Therefore,this thesis analyzes in detail the limitations of single point feature expression ability,and proposes an edge feature enhancement module and corresponding decoupling detection module.The backbone network scale reduction method,1x1 convolution dimension reduction and depth separable convolution module are used to compress the parameters.Through ablation experiments,comparative experiments,and real environment testing,the method proposed in this thesis has been proven to be well adapted to object detection tasks in indoor occluded scenes(3)This article designs and implements an application practice of object detection technology on a robot platform-an intelligent robot object finding system.This article designs a universal robot research and development platform for blocking the management of underlying hardware resources,and completes the system construction based on the platform.Users can remotely build maps and input target image information and category information through the Android APP.The robot server combines image retrieval technology on the basis of(1)and(2)to achieve precise indoor detection of designated targets.