Research On Object Segmentation And Recognition Algorithm Based On Kinect

Object recognition is a basic and important research field of computer vision,which is the basic study of image retrieval, video surveillance, robots and many otherapplications. So, the object recognition technology can widely be used.Since the subject of object recognition has been studied, great progress has beenmade, but the interference of background and light,image segmentation and shape hasalways been the difficulty of mutual occlusion of this subject. In recent years, with therapid development of image acquisition devices, in2010Microsoft launched athree-dimensional sensor Kinect sensor, which makes researchers easy to get the depthdata. In this paper, related work about object recognition using Kinect has beenstudied,depth of the image preprocessing algorithm, object segmentation algorithmbased on depth image and object recognition algorithm fusion of color information anddepth information have been studied. The main work of this paper include:Firstly, through research on the principle of Kinect how to get depth data andanalysis on the depth image of Kinect, the main noise in depth images was obtained,nearest neighbor interpolation method was used to repair the hole of depth image, andmedian filtering algorithm is used to filter the image edge.Secondly, the regional growth method is used to extract object region with thecharacteristics of Kinect depth image. And depth data is transformed into the pointcloud data into data, and the normal vector of each point cloud is calculated, the pointwhich belongs to the horizontal plane is determined by the direction of the normalvector, so the horizontal plane region of the depth image is extracted.Then, thesegmentation results of the horizontal area of the extracted region-growingsegmentation result of the correction, complete the object segmentation based on thedepth, Solve the segmentation of the object region and the horizontal area on the depthvalues caused by continuous over-segmentation problem.Finally, the model of the object recognition system has been introduced, theadvantages of Kinect that can provide a pair of color image and depth data have beenanalyzed, color histogram feature of the object and depth HOG feature of the objectbased on depth image were mixed to make research on the recognition algorithms. therecognition results that depth features and color features of objects were in differentweights were analyzed through experiments, and compared with the existing feature extraction methods. The results showed that recognition accuracy had been improved.