A Study On Real-time Target Tracking Algorithm Based On Event Camera

In contemporary society where intelligent driving and unmanned driving have become the hottest technology,and the research on target recognition and tracking of mobile robot as well as the visual SLAM algorithm has also become an extremely significant research direction.Since the current target detection and tracking technology is limited by the conditions of hardware and processor of the collecting equipment,there are still some problems existing in this technology,such as the recognition speed is not fast enough,the tracking accuracy is not high enough and it is more sensitive to environmental changes and other problems.A new type of sensor——Event Camera has been designed based on the principle of biological vision.The event camera has the characteristics of high dynamics,low data redundancy and low detention.Therefore,it is more suitable for real-time target detection and tracking system.The operational principle of an event camera is that when the pixels changes and exceeds a certain threshold after receiving the light intensity,the event camera will output the event information.If there is no light intensity change in the pixel dots or the light intensity change is well kept in the threshold range,the sensor will retain the previously recorded data,in which way,the real-time performance and computational efficiency of the system will be significantly improved.In this thesis,a target detection algorithm of the event camera based on MLS surface is proposed which aims to solve the problem that the event camera is not effective enough for target detection.The algorithm mentioned above processes the collected event information by utilizing the optical flow method based on the life cycle of events,and it uses the algorithm of random sampling consistency to fit the local plane,and puts forward the method of ML S surface fitting data,in which way,the amount of data required for plane fitting is reduced,and the smooth contour images obtained by regularization and edge thinning algorithm is added to ensure the accuracy of the results.This thesis presents a target tracking algorithm for event camera with high realtime performance based on improved FAST feature,aiming at solving the problem that the tracking speed of event camera is not fast enough.By improving the FAST feature detection algorithm,a self-adaption FAST feature detection algorithm is proposed to improve the anti-noise performance and robustness.In this thesis,the rBRIEF descriptor is used to describe the image features,so that the algorithm is entitled with rotational invariance,which enables the event camera to track the target well under high-speed rotation.In this thesis,the performance of the two algorithms is verified by experimental verification.The experimental results indicate that the target detection algorithm of event camera based on MLS surface has better performance of target recognition,and its accuracy and real-time performance are all improved.The tracking algorithm based on the improved FAST feature has been verified to have better robustness and real-time performance after it is carried out by several experiments in simple and complex,translation and rotation scenarios.