Vision Environment Perception Research For Service Robot Navigation

At present,with the dramatic increase in the performance of computer hardware and software,and the rapid progress in artificial intelligence technologies,unmanned vehicles and smart mobile robots and related fields have been rapidly developed.According to the current research situation,there are three core functional requirements of the intelligent robot can be defined as:action behavior,control ability and environment perception.At present,the focus and difficulty of robot research is the perception of environment.The environment perception is that the robot completes the perception and understanding of the surrounding environment through a series of sensors configured by itself,which can be divided into positioning,navigation,obstacle avoidance,and recognition.For example:using GPS to achieve high-precision positioning,using three-dimensional laser radar for three-dimensional space modeling,object detection,accurate distance measurement,using a gyroscope in real-time output of the robot's heading angle and acceleration and other information;use the camera to run image processing algorithms to identify real-time obstacle types and road segmentation,and take corresponding measures according to specific environmental conditions.The mobile robot can perceive the surrounding environment through visual recognition,calculate the road that can be driven,and calculate a predetermined travel route to achieve vehicle assistance or automatic travel.In the research of intelligent mobile robots,vision-based perception technology has always been a hot and difficult topic.Therefore,this paper based on the traditional visual image algorithm,and combined with the current research progress and development trend,provides a new set of visual environment perception solutions.The specific research content is as follows:1)In terms of the road surface recognition algorithms based on image processing,the relevant theories and concepts corresponding to the relevant image algorithms(e.g.filtering,edge detection,Hough transform,region growing,etc.)used on intelligent inspection robots are introduced.And for our special application situation,we propose a new road detection algorithm that is not affected by light.2)In terms of pedestrian detection,the relevant theories and concepts based on the HOG+SVM recognition algorithm are analyzed,and this algorithm is implemented and its speed and accuracy in pedestrian recognition are tested.3)In terms of the deep semantic learning algorithm for road segmentation,a deep learning model for campus road detection is first designed.Then photograph a large number of campus roads by ourselves and select some representative pictures to mark them.The marked pictures are used as training samples and sent to the model for training.The trained model is used for semantic segmentation of campus roads.The deep learning model we have trained not only has a high processing rate,but also achieves the current internationally advanced semantic segmentation accuracy.4)Through the relevant experiments and comparisons of these proposed pavement recognition algorithms,a set of intelligent patrol robots are finally provided to implement visual environment perception solutions for road recognition,semantic segmentation and pedestrian vehicle detection.The visual perception algorithm designed was tested on a mobile patrol robot developed by our laboratory and achieved good recognition accuracy.