Lightweight Recognition And Tracking Of Natural Image Based On Web

With the gradual penetration of AR services into all aspects of life,AR brings people unprecedented life experience and visual shock in many fields.WebAR,which runs on the browser side,enables users to experience AR applications by visiting web pages.Compared with the traditional AR based on app,it has the advantages of lightweight,cross platform and easy to spread.In the running process of WebAR,recognition tracking and rendering interaction are carried out at the same time.Due to the complex logic of AR algorithm,inefficient operation of JavaScript language and the limitation of browser performance,the traditional AR application on network platform has great limitations.At present,most of the AR applications running on the browser can only track the strong markers with high frame rate and attitude estimation.However,in the tracking process of natural images,there are often problems such as low frame rate and model jitter.How to bring richer and smoother experience to users has become a major problem of WebAR.The research direction of this paper is to implement lightweight and high performance AR system on front-end browsers.At the same time of tracking natural image,binding model position and posture and other AR functions,ensure that the system runs at a higher frame rate on the browser,and bring a smooth visual experience for users.Using WebAssembly technology to lightweight the OpenCV library,only encapsulate and compile the functional interface needed by the system into WASM module,so that the system can run at the original speed of bytecode in the front of the network in a lightweight way,and improve the operation efficiency;adding a special jitter optimization module,on the basis of using L-K optical flow method as a tracker to enhance the stability of the system,using weighted least square method and Kalman Filtering combination scheme is used to improve the least square using particle filter to get more stable pose data between images,and further compensate for the transition by Kalman Filtering,reduce stuck;aiming at adding other development modules to AR system,users can choose a model to interact with each other,enhance interaction between models and users,and enrich application scenarios.In order to verify the effectiveness of the above scheme,this paper designs a number of related experiments to test the functional modules of the system,and analyzes the performance improvement of the system in this paper by comparing the experimental data of the traditional scheme under the same conditions.The results show that this scheme not only realizes the functions of AR tracking and attitude estimation in browser,but also meets the requirements of low delay,high frame rate and real-time performance.