Research On Indoor Scene Recognition Of Home Service Robot Based On Deep Learning

The current level of service robot intelligence and autonomy still can't meet the demand of the human-robots communion.The robots can't directly understand the human's intention about scene because they lack the environmental cognition ability and the comprehension of the scene semantic object.The absence of the capability affects robot decision and behavior.How to narrow the cognitive gap between humans and robots in the same physical environment is the key to solve such problems.This question motivates indoor scene recognition become a research hot spot in the field of service robots.In the framework of depth learning,this paper combines the semantic information of the scene with the underlying convolution features to design a new scene recognition method.The main research results are as follows:(1)Scene Feature Description MethodThe object detection algorithm with deep learning is used to extract the feature of the object information in the scene.A data structure called the object vector is designed to associate the object with the category of scene.The prototype clustering algorithm is used to extract the category information from object vector to construct the prototype reliability vector.The deep convolution neural network model is used to classify and generate scene prediction vectors.Meanwhile,the original network structure is improved,and the global feature vectors of scene images are extracted by adding branches.(2)Scene Classification Method Combining Multi-scene FeaturesLogarithmic probability regression method is used to merge multiple description vectors to achieve fusion classification.Training on Places dataset and testing on another Indoor67 dataset to compare the accuracy of scene recognition results under different methods.(3)Recognition of Sequential Images from the Perspective of RobotsWhen the robot is working,the vision sensor obtains the sequence image of the scene.The CRF algorithm is introduced into the content processing of the sequence image,and the current scene category is predicted according to the content of the previous frames of the current frame.(4)Design of Robot Scene Recognition SystemIn this paper,a scene recognition system for home service robots is designed to realize real-time recognition of home service robots in real scenes.