Research On Regional Semantic Map Building And Object Search Method For Mobile Robot

In order to make the robot work better in human-centered complex indoor environments,it is necessary to develop the service robot's ability to understand the surrounding environment semantics,which exceeds its traditional ability to avoid obstacles,autonomously navigate,or build maps.Based on the understanding of the semantics of the surrounding environment,service robots also need to perceive and identify common living objects in human life,so as to find life supplies(and subsequent visual grasping,handling,and operation)for the elderly and disabled people with mobility problems.This paper focuses on the above issues,and focuses on the regional semantic mapping method,object recognition method and indoor visual search method for service robots.Firstly,this paper studied a semantic topological map construction method based on deep learning.According to the characteristics of different indoor scenes,this paper uses a deep convolutional neural network to classify them separately,and combines the laser data and visual data to estimate the semantic region,then builds a semantic map layer on the basis of a two-dimensional grid map.Further,in order to express some key areas in the indoor environment,this paper subdivides each indoor semantic area into several functional areas on the basis of semantic maps,and inserts each functional area into semantic maps in the form of semantic topological nodes,thus constructing a semantic topology map.Secondly,an indoor object detection system based on deep learning was constructed.The system is based on the current popular SSD(Single Shot Multi Box Detector)method in the object detection field.This method discretizes the output space of the bounding box into a series of default boxes with different aspect ratios and scales at each feature map position.In the prediction,the network predicts the category and confidence of several object targets in each default box,and adjusts the rectangle to match the position and size of the object.Thirdly,this paper proposes an indoor object search method based on semantic maps.This article assumes that a specific object type is located in a certain semantic region of the room.When the robot performs an object search task,it first determines the target area to reach according to the correspondence between the label of the object target and the semantic area.Then according to the topological relationship between the semantic region where the robot is currently located and the target semantic region,the path is planned at the semantic level and the grid level respectively.When searching in the target area,the work is performed according to whether or not the robot has a priori information on the distribution of the target object in two cases,and an object detection system is used to identify the target object in the search process.Finally,Based on the omnidirectional mobile robot experimental platform,this paper conducts experiments on the regional semantic mapping method proposed in this paper,the target identification method,and two search methods for indoor targets,and analyzes the experimental results in the actual environment.