Research On Road Classification Method For Mobile Robot

Mobile robot is an important branch of robotics research area which has been widely applicated in many fields. It always needs to change different control strategies to adapt variable lane types when it moves automatically. Therefore, it is one of the key technologies that how to identificate road types autonomously. Nowadays, it is still a difficulty that how mobile robots classify lanes type quickly under variable environment. The road classification method oriented to both indoor and outdoor mobile robots has been studied based on machine learning in this thesis, and the main contents and conclusions are as follows,(1) Orient to indoor mobile robots, support vector machine algorithm, compared with naive Bayes, is employed to classify indoor preinstalled lanes. It is proved by experiment that the results obtained with the support vector machine method are better when the center-line, the left line and right line are taken as input features.(2) An implementation method based on wireless network is proposed in order to use support vector machine under limit hardware resources circumstance, the support vector machines are trained in the SaaS (software as a service) providers using this method, and then microcontroller gains or updates models to predict new samples from SaaS providers through wireless network. The time complexity of using support vector machines in microcontrollers is analyzed before verification experiments. The experiments show that the method is good and fit for indoor mobile robots.(3) A topological map editor is programmed using QT application based on road network definition file (RNDF) format and the information about lane types is included in the way point attributes. The editor can match up satellite images with GPS data, build topological maps in RNDF format and update the way point attributes such as lane type manually. The result shows that the map builded by the editor meets the requirement of mobile robots.(4) The support vector machine, aim at the problem that the lane type can only be tagged in the editor manually, is applied to classify road types automatically using video images. The video images are captured at the same time with GPS data. The GIST feature of images is used to classify the traffic scenes under single-frame circumstance. The results show that it has high accuracy using support vector machine to classify video images.