Navigation And Object Detection Algorithms For Inspection Robot In Industrial Environments

For complex industrial environments,most of inspection tasks are carried out manually.These tasks are time-consuming and require certain specialized skills.Meanwhile,for high-risk industrial environments,the safety must be ensured for workers.In this thesis,we design an intelligent inspection robot for complex industrial environments,and study the key issues arising in navigation,object detection and recognition.Our contribution includes:· Robot Localization: Based on the Monte Carlo method and KLD sampling strategy,we designed and implemented a environment positioning algorithm for inspection robots.The algorithm uses the real-time lidar data and the environment map for matching,so that the robot could perform precise positioning and safe navigation in known industrial scenes.Besides,an indicator is proposed to evaluate the localization quality of the algorithm.· Risk Detection and Avoidance: Considering the complex terrain and obstacles that may exist in the industrial environment,in order to ensure the safety of the inspection robots,we designed and implemented a dangerous environment detection algorithm based on monocular vision.With the help of deep learning technology,we use ResNet to classify dangerous environments and use edge computing to deploy the model on the robot.At the same time,we proposed four methods for data enhancement,and apply the methods to the production of data sets.· Meter Recognition and Reading: Aiming at the inspection tasks of industrial inspection robots,we designed and proposed an instrument detection and recognition based on computer vision and deep learning.With the help of MobileNet v3 and self-made data set for training,the target meter position could be extracted from the image.The target meter image is corrected based on homography,and image matching is used to complete the tasks of pointer recognition and meter reading.· Software and Hardware Architecture: Based on the differential wheeled sports chassis,we designed the hardware system of the inspection robot,combined with IMU,lidar,wheeled odometer,vision sensor and other modules.All the sensors are used to build a multiple-sensor communication framework.The inspection robot is based on ROS,OpenCV,Pytorch and other software platforms,combined with the above algorithms to realize the inspection task in the industrial scene.