A High-efficient Method On Object Search Of Service Robot In Unknown Family EnvironmeMt

In family environment service robot object search is an important basis and premise of carrying out tasks such as object delivery and grabbing,which is of great significance to improve of robot intelligence and service ability.At present,most of researches on this topic adopt constructing a semantic map of the environment for object search.But because of family differences a new semantic map need to be constructed repeatedly to obtain object information in an unknown family environment,which reduces the efficiency and adaptability of the robot object search.To solve this problem,some researches on object search of service robot has been done in this paper,whose application background is unknown family environment.A high-efficient method on object search of service robot in unknown family environment is proposed,which utilizes object search prior knowledge as guidance and family scene recognition model and object detection model as tools.The robot finds a target object in a family object search knowledge base to get scene and object knowledge of the target object.Under the guidance of the scene knowledge,the robot uses an indoor scene recognition model to recognize target scene so as to reduce the search space.In the target scene,under the guidance of object knowledge the robot completes the object search task with the object detection model.The contents and innovation of this paper are as follows:1.The family object search knowledge base is constructed,which can provide object co-occurrence knowledge and object scene co-occurrence knowledge for the service robot object search.The knowledge base can not only improve the efficiency of object search in an unknown family environment,but also meet the requirements of object search in different family environments and improve the adaptability to the environment.In the knowledge base the prior knowledge of object search is generated automatically based on graph model and probability statistics by using the annotation information of a large-scale dataset and does not need to be added manually.It covers 358 common household objects,which can meet task requirements of object search in most household environment and has a function of knowledge update.2.To take advantage of the indoor scene knowledge to reduce the search space and improve the eff-iciency of object search,this paper proposes an indoor scene recognition model based on transfer learning and CNN feature matching,solving a problem that indoor scene data are too small to train a scene model and the recognition accuracy is too low.The model only use a small amount of indoor scene data can get high accuracy,which uses transfer learning method to generate scene category features based on CNN features of indoor scene images.The scene category features are matched with CNN features of a test scene image to obtain scene recognition results.The proposed model is evaluated on two public scene datasets Scene 15 and MIT 67,obtaining recognition accuracy of 96.49%and 81.69%respectively.Compared with other advanced methods,our model has advantages on both recognition accuracy and speed.3.An object detection model based on the priori knowledge is proposed to solve the problem of small object detection in the final step of object search.Using object detection prior knowledge,an object detection model is built based on ResNet in order to improve the detection performance.To improve the detection ability at the same time,an instance object detection model is proposed,with which the robot can detect individual objects.solving the problem that current object detection models can't realize instance object detection.Experimental results show that the model has good performance on the detection of small objects and instance objects and improves the ability,efficiency and level of robot object searchThe object search method proposed in this paper,closely combining family object search knowledge base,indoor scene recognition and object detection.has good practical value and significance on improving the efficiency,ability and adaptability of service robot object search in unknown family environment.