The Design And Implementation Of The Object Recognition Engine Based On Opencv

In theory research, the object recognition has been popular in the field of computer vision research in the last decade. There are a lot of relevant work in this direction in top conferences and journals each year. Some scenarios detailed in such as face recognition and fingerprint recognition has been mature application. In this paper, research object recognition and designed and implemented an object recognition engine based on OpenCV, to try to solve some specific object recognition needs. The main work completed is as follows:(1)Object recognition plan based on features:Firstly, the research is about the object recognition algorithms, which mainly consists of two parts:features detection and features matching. Study the principle and implementation process of three features detection algorithm in detail, Harris feature, FAST feature, SURF feature, and analyze the characteristics of the algorithms. Understanding of the feature point detection algorithm in the practical application helps select the appropriate algorithm for a particular scene and optimization. For object recognition, also need to match the image features with sample library, then obtain the corresponding tag information. Such matching feature are also an important part. In the paper compares the linear search algorithms, FLANN and some filtering algorithms of matching pairs. The matching algorithm with some combination of filtering algorithms designed two matching plans.(2) Design of persistence of sample:Persistence is an essential part of a particular engine. The sample data persistence preservation is an important part of practical application, and to serialize related features information and tag information. After research and analysis, use leveldb as a storage engine, persistence saved the serialized information related tags and some metadata.(3) Design and Implementation of object recognition engine:By analyzing the demand for content, using a modular design concept, design and implementation of the entire object recognition engine, and design an efficient high-speed matching iterator to traverse sample library. At last, test the relative performance of the entire engine.