Indoor Scene Recognition Based On Image Segmentation And Feature Extraction

With the continuous improvement of people's living standards,home service robots gradually come into people's lives and are receiving more and more social attention.The understanding and recognition of the indoor environment by home service robots is a key factor for the home service robots to serve human beings better and execute human instructions accurately.Image scene recognition mainly focuses on organizing,analyzing and reasoning the global semantics of ingested scene images to obtain the semantic information of its own scene images.It has always been a hot spot for people to study.Inthis paper,we study indoor scene recognition,the specific research contents are as follows:1.A modified algorithm is developed based on Mean Shift segmentation method,fully taking into account the effective control of segmentation scale and the scene depth information of segmentation process.In order to solve the problem of controlling segmentation scale,the concept of edge sensitivity is proposed so that the control ability of algorithm scale block is improved.Two methods of getting the depth information based on binocular stereo matching and image sensor Kinect are employed to deal with depth information fusion.Both the methods realize the fusion successfully and the performances of segmentation are enhanced obviously.The results of experiments demonstrate that our proposed method is more preponderant than traditional algorithms.Our proposed algorithm could not only control segmentation scale,but also split the special situations that the traditional algorithms could not deal with(such as segments with similar color close to each other).2.After the image is divided into blocks,the eigenvectors are extracted and the whole image is analyzed and understood based on segmented blocks' features.The color features,HOG features and shape features of the segmented blocks are respectively extracted to prepare for the subsequent image recognition and classification.3.The Bag-of-words model is used to integrate the low-level features of the image to obtain the semantic content of the image.The traditional Bag-of-words model takes the SIFT feature as the visual feature.In this paper,the segmentation features and the SIFT feature constitute the visual features of the image,and the two visual vocabularies are formed by using K-means clustering algorithm.Based on two visual vocabularies,an expression vector of the image is formed.4.Taking the laboratory four rooms as the research object to complete the scene recognition experiment.After the collected scene images are spliced to form a panoramic image of the scene,the panoramic image is taken as a unit to complete the scene recognition.Experiments show that the Bag-of-words model which combines the features of the segmented blocks performs scene recognition,which has a higher recognition rate than the Bag-of-words model which uses the SIFT feature or the segmented block feature as the visual word alone.