| Obtaining detailed information about objects around the work scene is the basis for indoor service robots to carry out motion planning and complete specific tasks,which is the key to realize the intelligence of indoor service robots.The acquisition of class,location and distance information of objects in the working environment allows the indoor service robot to respond appropriately to the obtained target information.In order to enable the indoor service robot to realize the function of acquiring the category,location and distance information of specific targets simultaneously,this paper conducts in-depth research on binocular stereo vision technology and target detection technology,and designs and implements an indoor target detection and ranging system based on binocular vision that can run on an embedded system by combining the working scenarios and practical application requirements of indoor service robots.The main work of this paper is as follows:1.Based on the imaging of the camera and the principle of binocular stereo vision ranging,combined with the needs of binocular stereo vision ranging,choose a suitable camera calibration method to calibrate the binocular camera,and use the calibrated binocular camera parameters to stereo correct the binocular image to meet the requirements of the use of stereo matching algorithm.2.The applicability of binocular stereo matching algorithm for application scenarios is studied,and the advantages and disadvantages of different stereo matching algorithms are analyzed.In order to improve the anti-noise capability of the stereo matching algorithm and ensure the applicability and accuracy of the algorithm in specific application scenarios,targeted improvements are made to the stereo matching ranging algorithm based on the Census transform.The main improvements are: optimizing the calculation method of Census transform and selecting the reference value of Census transform by setting a threshold to improve the noise resistance of the algorithm;incorporating the image gradient information in the initial matching cost to improve the stereo matching effect;reducing the cost aggregation path to improve the operation efficiency;optimizing the selection of the ranging disparity to improve the reliability of the ranging.3.the target detection algorithm is investigated,and a suitable lightweight target detection algorithm is adopted in this paper for the real-time detection and ranging needs of indoor service robots,taking into account the implementation difficulty and operational efficiency of the whole target detection and ranging system.A new dataset is produced by screening common indoor target species from the open source MS COCO dataset,training the target detection model and evaluating the model.4.Integrate the system to achieve acquiring target location and category while acquiring distance information,complete embedded deployment and perform experimental analysis.Object detection functionality is implemented through the Open CV call model,reducing environmental dependencies when deployed in embedded devices.Simulate the working scenario of indoor service robots to experimentally verify the target detection effect and range accuracy of the system,and test the operation efficiency of the system on the embedded devices.The experiments show that the system designed in this paper achieves the predetermined design goals,and the m AP of the model obtained by training using the produced dataset is 53.8%,the ranging error within5 meters is less than 6%,and the average operation speed on the embedded device is up to 27 FPS,which meets the requirements of the indoor service robot for indoor scenario applications. |
