Design And Implementation Of Intelligent Recognition System For Indoor Scenes

With the emergence and development of deep learning,the study of machine vision has become a hot topic in domestic and foreign scholars,among which the recognition and classification of indoor environment scene has made great progress.The development of indoor environment scene recognition and classification technology has a basic and constructive effect on the follow-up research of the Family mobile robot,and it also has important significance for improving the quality of life of the citizens.The traditional scene recognition and classification technology adopts artificial extraction feature,which has many disadvantages,such as low efficiency,poor expansibility,poor real-time performance and low accuracy,therefore,a new network structure recognition model is proposed,which realizes the intelligent recognition system for indoor scene of video image.Based on the research of convolution neural network,this paper presents a convolution neural network structure Place-AlexNet with Bayesian filter and random forest classifier,and designs and realizes an intelligent recognition system for indoor scene.Image Acquisition module,the system selects 11 classes in place205 image data set as the training dataset of scene recognition model,and realizes the acquisition of test image data with visual sensor.The image preprocessing module performs bilinear interpolation on the image and normalizes the operation,and the image is unified to the size of 227*227.The image is averaged,the data of each dimension is centralized to 0,and the amount of network calculation is reduced,and the preprocessing is completed.Scene recognition Model building module,which is based on the AlexNet of 5 convolution layers and 3 full connection layers,embeds the Bayesian filter in the last full connection layer to realize the time continuous scene recognition function of video image,and uses random forest as classifier to realize the classification of scene.Using the reverse propagation cycle of network structure to train the updating weights to reduce the errors until the desired results are achieved.Visual Display module,this paper uses the visual software Rviz of the robot operatingsystem to realize the visualization of the indoor scenes intelligent recognition system.This system uses the Python language on the Ubuntu platform to program the interior scene recognition system to compile,after the system test,has proved that uses the Place-AlexNet network recognition model to the indoor environment scene recognition question validity and the stability,and the correct rate of recognition on average can reach 88.23%,it has realized the expected requirements of the system.