Asynchronous Event Feature Detection And Tracking Technique Based On Event Cameras

Recently,with the rapid development of vision sensors and computer vision technologies,visual SLAM has been widely studied and used in many fields in academic and industrial fields,such as mobile robotics,virtual reality and so on.However,standard cameras suffer from several technological limitations,such as high energy consumption,high latency,high information redundancy,and low temporal resolution,which prevent their utilization in high speed robotic and visual applications,for example,self-driving cars,unmanned aerial vehicles,etc.In addition,due to the complexity(high dynamic range,etc.)of practical application scenarios,the existing visual SLAM systems based on standard images for simultaneous localization and mapping tasks will inevitably lead to the problem of algorithm failure,which seriously affects the robustness and stability of the systems.With the development of neuromorphological imaging technology and silicon retinal technology,dynamic vision sensor,namely event camera,began to appear and gradually commercialized.Event cameras capture dynamic changes in the scene based on event-driven method,and respond to pixel-level brightness changes resulting asynchronous event streams.This new type of dynamic vision sensor has the advantages of low latency,low energy consumption,high temporal resolution and high dynamic range.It has the potential to meet the above challenges.Based on the event camera,this paper mainly studies the method of processing the asynchronous data output from the event camera,aiming to design and implement a new asynchronous event feature detection and tracking technique.The research focuses on asynchronous event feature detection technique,event feature gradient descriptor construction technique and asynchronous event feature detection and tracking model.Specially,the main contributions of this paper can be summarized as follows:(1)Aiming to extract features from the event streams of the event camera,we propose an asynchronous event corner feature detection method directly operating on the asynchronous event streams,It includes two core strategies: global SAE maintenance and update strategy,candidate event corner feature selection and refinement strategy.(2)Aiming to describe the event feature using the information of event streams,we present a gradient descriptor construction method for event features,The constructed gradient descriptor contains the local gradient distribution information of event corners,which can help to achieve quantitative measurements of the similarity between event feature pairs.(3)Based on the above research,we design an asynchronous event feature detection and tracking model for event cameras adopted with an asynchronous event feature matching algorithm,The input of the model is the asynchronous event steams,and the output is the tracks of the event features,It includes three stages:asynchronous event corner feature detection,event feature gradient descriptor construction and asynchronous event feature tracking.Based on the above research,we design and implement a prototype system for asynchronous event feature detection and tracking with ROS as data communication interfac.And we evaluate the proposed prototype system on the public datasets.The experimental results show that the system can run in high dynamic range scenarios.The evaluation demonstrates that our proposed algorithm performs better in terms of tracking accuracy and real-time performance when compared with the state-of-the-art asynchronous feature tracking methods,and with no compromise on the feature tracking lifetime.The above asynchronous event feature detection and tracking technique directly operates on the asynchronous event streams,and fully exploits the asynchronous characteristics of event cameras.The technique can adapt to high-speed motion,high dynamic range and other challenging scenarios to a certain extent.It provides a new research idea for solving the robustness of visual SLAM in the challenging scenarios,and provides support for the construction of complete visual odometry system and SLAM system.