Research On Three-dimensional Object Recognition Based On Binary Descriptor

3D object recognition is a very challenging research area in computer vision. Existing algorithms for 3D object recognition can broadly be classified into two categories: global feature based and local feature based algorithms. The global feature based algorithms construct a set of features which encode the geometric properties of the entire 3D object. The global feature based algorithms are sensitive to occlusion and clutter. In contrast, the local feature based algorithms are robust to occlusion and clutter. They are therefore even suitable to recognize partially visible objects in a cluttered scene. Therefore, this thesis only focus on the local feature based algorithms. Since the performance of existing local feature based algorithms are limited by the computing and matching efficiency of feature descriptors, the local feature based algorithms are difficult to implement in real time. In order to solve this problem, the main research works carries out in this thesis are as follows:① This thesis provides a survey on the current state of 3D object recognition research. Firstly, a comparative summary about datatypes and datasets of range data are reported in this thesis, then the principles of range data imaging are explored. Subsequently, existing 3D object recognition algorithms are divided into two broad categories: global feature based and local feature based algorithms. The merits and limitations of two categories of algorithms were analyzed, and then the research direction is focus on local features based algorithms. Finally, typical local feature based algorithm is divided into two phases: 3D feature extraction and surface matching, and then surveyed related methods of respective phases.② This thesis analyses the construction principles of three feature descriptors, which including: spin image(SI), signature of histograms of orientations(SHOT) and rotational projection statistics(Ro PS).③ This thesis proposes a binary keypoint descriptor and constructs a 3D object recognition algorithm. This thesis proposes a robust, fast, lightweight binary descriptor, which improves the computing and matching efficiency. Our constructed 3D object recognition algorithm has good compatibility, and uses to measure the recognition rates of various descriptors on the standard database.④ This thesis designs experiments to test the performances of the proposed binary descriptor and 3D object recognition algorithm. Experiment results shows that the matching efficiency of the binary descriptor significantly higher than SI, SHOT and Ro PS. This thesis uses the constructed recognition algorithm to measure the recognition rates of the binary descriptor and Ro PS on the UWA standard database.