Research On Semantic Map Building For Mobile Robot Based On Multiple Feature Fusion

In recent year, with the development of intelligent mobile robot, it’s an urgent problem facing related researcher that how to make robot obtain more environment information and deal with complex service tasks. Semantic map, which has been paid more and more attention by researchers both at home and abroad, contains rich environment information, and help the robot to complete service tasks efficiently.In order to complete the service task, a multiple feature fusion method is proposed to construct the semantic map which has more semantic information than traditional map. Firstly, according to the laser data, robot gets the environment structure and builds two kinds of maps. An Immune algorithm and antibody mutation mechanisms is provided to plan path. Then detection of objectness is introduced to offer a candidate window. On this basis, candidate window can be classified by HOG feature. Finally, a region segmentation algorithm based on voronoi diagram is used to divide topology node control region. The main work has the following aspects:1. Passable space construction based on tree structure.With the high time complexity of admissible space construction, an algorithm based on tree structure is proposed to solve this problem. According to the proposed algorithm, section, as an infrastructure, describes passable space in each layer of laser data and is traversed to convert to tree structure so as to realize rapid modeling of the environment.2. Path planning based on antibody mutation.Through immune algorithm, passable node is evaluated. On this basis, the local optimal passable node is selected to improve exploration efficiency. Immune algorithm reduce the time of backtracking but it also lead to the robot turn around on the spot frequently to reach the passable node. A path planning algorithm based on antibody mutation is proposed to addressed the above problem. The position of passable node will be adjust to avoid pivot steering.3. Score correction based on region similarityIn order to avoid the false result in object detection caused by poor location of the candidate window, a score correction based on region similarity is proposed to synthetically consider with color, texture, and the size of candidate window. In this process, a fusion algorithm based on color feature is introduced to enhance computing efficiency. By using the proposed method, we can optimize the selection of candidate windows, and lay the foundation for the next object recognition.4. Hierarchical model construction based on multiple feature fusionRestricted by the sensor capability, the environment information collected by a single sensor mobile robot has certain limitations, and will lead to simple environment model. Therefore, in the semantic mapping process, the environmental features collected by various sensors need to be fused to enrich environment map. A hierarchical map model is proposed to achieve this purpose. It divide the map into the geometric layer, topological layer and semantic layer. Furthermore semantic map which contains rich environment information is constructed.The validity and practicability of the proposed semantic map building approach is validated by a lot of experiments in an actual environment. The semantic map building with above method can help robot to complete service task, and have a certain theoretical and practical value.