Research On Indoor Object Detection Based On Deep Learning

With the continuous development of image analysis technology,the object detection has attracted more and more attention from researchers.Scene information as an essential information in the image is a vital information for the object detection.The scene is mainly divided into indoor scenes and outdoor scenes.The indoor scene is closely related to human life,so the indoor scene object detection as an important branch of target detection has been widely concerned and studied.The target detection of indoor scene can make great contribution to people's life and work.This paper studies a new indoor target detection method based on the existing general purpose detection framework.The main research contents of this paper are as follows:1.This thesis proposes an indoor object detection network based on recurrent convolution neural network.Firstly,the global information module is constructed by using recurrent convolutional neural network to construct the global information module,and then it is combined with the existing detection network model to detect the indoor scene.The detection network model proposed in this thesis can make full use of the contextual information of images to improve the object detection effect of indoor scenes.2.In this thesis,the overlapping target detection algorithm based on confidence adjustment is studied.Using the non-maximal suppression algorithm to filter the detection box,the detection box of some overlapped targets is filtered out.Based on the overlapping area and detecting box category adjust frame's confidence,and then object box is obtained by confidence threshold filter detection box,to the detection effect of overlapping target can be effectively improved.3.In this thesis,an optimization method of object detection based on indoor object attribute is proposed.By using the coexistence relationship between indoor targets,the probability of them is statistically obtained,and the confidence level of the detection box is adjusted according to the probability of the transformation factor and the hard interval.The optimization method of this paper can effectively reduce the appearance of false and omission.In order to train the network model of this thesis and verify the validity of this method,this thesis constructs an indoor target detection database.Both objective experimental data and subjective experimental results show that the network constructed in this paper and the algorithm proposed in this paper have a certain improvement in the indoor object detection effect.