Research On Indoor Target Recognition Technology Based On KINECT

Indoor target recognition is a process that a target is separated and recognized from other targets in indoor environment, which has important significance to the research of the home service robot and the design of home monitoring system. Because the targets are various and the light condition is not stable in indoor environment, these interference factors often affect the accuracy of target recognition. How to overcome the interference factors and get a better recognition effect is a problem that needs to be solved. In this paper, a recognition method, based on Kinect color and depth image fused features of indoor targets, is proposed, including the research of image preprocessing and target extraction, feature extraction and feature fusion and the classification of targets.The main work of this paper is as follows:(1) Image preprocessing and target area extraction of indoor target. Firstly the color and depth images of indoor targets are collected by Kinect. With the regard of the problem of the obtained depth images' holes, a modified median filtering method is proposed to repair the holes. With the regard of the noise interference problems of indoor target images, this paper uses mean filtering method to remove the noise. With the regard of the problem of target region extraction, the depth image edge detection method is used to extract all kinds of objects in the indoor image.(2) Features extraction and features fusion of indoor targets. Image features are various and each feature has its own characteristics. This paper analyzes the extraction principle and advantage of the color feature, shape feature and texture feature. Then the paper chooses the gray statistical characteristics, moment invariant features, wavelet texture feature as the extraction of image features according to the practice of target images. Features fusion can improve the efficiency of target recognition, so this paper analyzes the principle of the pixel-level fusion, feature-level fusion and decision-level fusion method and uses the improved PCA method to fuse features data in feature-level, and then get a better separable feature as the classification feature.(3) Classifier selection and classification experiment. The performance of the classifier has an important influence on the accuracy of target recognition. Support vector machine has a unique advantage in the case of small sample data. In this paper, support vector machine is chosen as the classifier, the basic principle of support vector machine is analyzed firstly, and then the parameter optimization of support vector machine is studied. Genetic algorithm is chosen to find the best parameter based on the analysis of the advantages and disadvantages of grid search algorithm, genetic algorithm and particle swarm optimization algorithm. The classification experiment of several indoor targets is carried out. The results show that the classification rate of fusion feature reaches to 99.3%, which is better than other image features and gets better recognition results.This study uses Kinect sensor to research the classification question of the indoor targets, the fused features of color image and depth image are used as the classification feature, which gets a better recognition effect and provides an important reference for the design and development of indoor intelligent recognition system.