Indoor Object Detection Based On Machine Vision

The perception of objects in indoor environments is one of the capabilities that indoor mobile robots must possess.Object detection is an important role for robots in human-computer interaction,navigation,path planning,and motion control.This article's algorithm that is a new object detection algorithm are based on existing algorithms.The algorithm uses neural network to extract candidate regions and classifies them using multi-feature fusion + PCA + SVM algorithm.This article focuses on indoor object detection.This paper researched indoor object detection,and its main research work is:(1)The whole connection layer of the VGG16 network model was modified.Using the modified network structure to train network model.The features of the fc7 layer as CNN features are extracted by the training model.Because the model trained by VGG16 network on ImageNet has been published,this paper uses the parameters of the convolutional layer in the public model to initialize the network.Because the model trained by VGG16 network on ImageNet has been published,this paper uses the parameters of the convolutional layer in the public model to initialize the network(2)Combine CNN features with the traditional features such as HOG features and LBP features to feature fusion.PCA was used to reduce the dimension,then the classification model was trained by SVM classifier.The CNN+HOG+LBP+SVM algorithm proposed in this paper gets a good performance in indoor object recognition,which is superior to the ordinary CNN model and the traditional classifier.(3)Researched the influence of parameters of PCA parameters and multi-feature fusion on experimental results,and select the best parameters through experimental comparison.This article uses two data sets,one is the VOC2007 data set,and the other is the ImageNet1(some indoor object pictures in the ImageNet data set).In the experiment,six object detection algorithms were used in this paper.And those methods were compared.