Research On Target Searching Method Of Mobile Robot In Unknown Environment

Robotics embodies degree of science and technology development of the contemporary era, which is named as "most significant automation", involves many disciplines and covers a great number of research fields. The big demand for mobile robot intellectuality become increasing under its widespread applications. People expect mobile robots to deal with complex, dynamic and unknown environments, and could be capable for autonomous environment exploration and target searching tasks. In the research of mobile robot target searching methods and theories, many achievements have been reached in known environments, but for unknown environment, there has not established a unified and complete architecture, as yet plenty of critical theories and techniques are needed to be improved and resolved. In this dissertation, the complexity conception coming from the human search experience was introduced in the mobile robot target searching task, and a novel target searching method based on scene complexity was proposed.First of all, formalization definition of scene complexity was given by analyzing the main factors relative to scene complexity, which combined image complexity with laser ranges complexity. In order to calculate scene image complexity, texture feature, salience region and edge ratio corresponding to the image were calculated respectively, then these factors were integrated together as normalization values; for the calculation of laser ranges complexity, the coefficient of variation and mean value corresponding to laser ranges data were integrated as complexity evaluation factors and then normalized.Secondly, an environment exploration method based on scene complexity was presented. In this method, the next exploration scene was selected by comparing several scenes complexity, and the next exploration position was determined by its depth information. Meanwhile, to solve the SALM problem, a novel SLAM algorithm based on particle filter was proposed, in which an adaptive resample method was employed to improve the resample performance.Finally, by combining environment exploration and object detection, the target searching method based on scene complexity was realized. One more factor named as "object likelihood" was added into the scene complexity, which could make the search process more clear and definite. In object detection process, an object detection method based on Harris-SIFT feature was presented, in order to solve the divergence phenomenon of match points and the issue of object location.The experimental results show that the calculation method of scene complexity is in accord with subjective perception of human, which reflects the scene intrinsic characteristics. The proposed environment exploration method and target searching method are effective and robust, and different complexity thresholds lead to different granularity of exploration or searching results. The proposed improved SLAM method can improve the performance and decrease particle degradation effectively.