Simultaneous Recognition And Modeling Of Indoor Scene Objects

Service robot has been playing a very important role in people’s lives. With the continuous improvement of people’s living standard, the service robot needs and requirements are rising, so how to recognize objects and establish the object models in an unfamiliar indoor environment for robot becomes very important. It directly affects the control of the robot. We propose a basic framework to address this hot issue. The main questions focused on 1) how to recognize objects in a complex indoor environment 2) how to establish objects models which retaining the three-dimensional object information and providing better recognition for the next time.It is difficult for service robot to recognize objects and establish the object models in a complex indoor environment. It mainly performances in algorithm’s robustness and timeliness, and the timeliness performances in object recognition rate and model establishment rate. In this paper, we do deep research on the above two issues and propose an efficient framework to recognize objects and establish object models. The main work includes the following aspects:1.We classify and summary the object recognition and modeling algorithm and introduce the research status of object recognition algorithm.We overview the object recognition algorithm from 2D and 3D, mainly introduce the 3D object recognition and pose estimation method and object modeling and data storage method. We analyze the advantages and disadvantages of these methods in detail.2. Before the object identification and modeling, you need to separate objects from the background. In this paper, we overview the 2D object segmentation algorithms such as a salient segmentation algorithm and 3D object segmentation algorithms such as RANSAC algorithm, clustering segmentation algorithm. We need to remove the human body’s interference before object detection, so we propose a salient detection method applied to the human body segmentation in static images to remove the human body, and then we do indoor object segmentation.3.In indoor object recognition and modeling proposed by Alvaro Collet Romea et al, we find the method’result has a low recognition rate and the method can not recognize object online. In this paper,we merge the 2D SIFT features and 3D FPFH features into the framework proposed by Collet Romea et al. At last, the improved method not only increases the recognition rate, but also increases the callback rate.4. In order to improve the online object recognition and modeling speed, we propose an improved object modeling method. The method merges two views which pose is close each other together and compresses every node’s data. Object Modeling is not only to display the three-dimensional structural information of the object, but also to serve the object recognition. Effective data extraction and low redundant information not only can reduce the storage pressure, but also can improve the speed of object recognition.In this paper, we do simulation experiments to verify the method mentioned in the paper with the help of image processing library OpenCV and PCL. The results show the effectiveness of the improved method. The end of the paper is the conclusion which describes the innovations and key results and points out future research directions.