Motion Compensation For Event Cameras

Event cameras are biologically inspired sensors that can capture brightness changes at an extremely high speed and output a series of ”events”,which record the time,location and polarity of the brightness change.Compared with traditional cameras,event cameras have some advantages: high dynamic range,high temporal resolution,low latency,and low power consumption.Therefore,event cameras have great potential in challenging environments such as high speed and high dynamic range.However,because of the output characteristics of event cameras,new approaches are needed to unlock the potential.At present,a common method is to create images by accumulating events over a short period of time,and then apply traditional image-based algorithms.In the process of frame construction,an important part is motion compensation,so as to eliminate motion blur caused by camera motions.The motion models used by the current motion compensation algorithms are all linear,which are not applicable to the case of a long time interval and rapid change of camera speed.This thesis presents a motion compensation algorithm based on nonlinear motion model,which avoids the linear motion assumption in a very short time in existing algorithms.The proposed method sets motion parameters as a nonlinear function of time,and use the energy maximization framework to solve the problem.In order to prevent noise,the proposed algorithm adds a smooth constraint on motion parameters.Finally,this thesis gives the optimization process of the algorithm and analyzes the computational complexity.In order to better evaluate the performance of different algorithms,a new evaluation index named normalized energy is proposed in this thesis.This index mainly depends on the complexity of scene texture and motion compensation effect.Finally,the parameter setting of the algorithm is analyzed,and the performance of the algorithm is evaluated by qualitative and quantitative experiments on public datasets and self-built datasets.The experimental results show that the proposed algorithm not only has good anti-noise performance,but also has better motion compensation effect.