Research On Target Tracking Algorithm Based On Dynamic Vision Sensor

Dynamic vision sensors are bio-inspired sensors that work radically different from traditional sensors.Compared with traditional vision sensor,it has the characteristics of low delay,high dynamic range,small data redundancy,low power consumption and working asynchronously.The dynamic vision sensor has no motion blur and is suitable for collecting high-speed moving targets.In addition,the dynamic vision sensor can achieve better data collection effects in reflective scenes and dark environments.In view of these advantages of the dynamic vision sensor,it has great potential for tracking.The purpose of this research is to explore the application of the dynamic vision sensor in tracking and to release the potential of the dynamic vision sensor in target tracking.In order to achieve this goal,my work focus on three aspects.Fisrt,feature detection based on dynamic vision sensor.Secondly,feature tracking using frames and events.Finally,feature tracking based on optical flow detection.The research content includes the following parts.First,in order to detect feature,this research studies the feature detection algorithm of dynamic vision sensor.First,this research studyies the feature detection algorithm based on integral image.The integral image is generated using event stream,and then feature detection is achieved by basic feature detection algorithms.In addition,the dynamic vision sensor work asynchronously.According to the characteristics,an asynchronous feature point detection algorithm is studied.Then,in order to take advantage of the low latency and high time resolution of the dynamic vision sensor,my work studies the tracking algorithm combining the gray image and the event stream.This algorithm generates integral image by event stream integration,and generates prediction image by grayscale graphs,and uses the relationship between the two to find affine transformation parameters to achieve tracking.In addition,for the feature tracking of dynamic vision sensors,a tracking result evaluation method is designed based on the KLT algorithm.The tracking algorithm achieve high accuracy feature tracking,there is no time dead zone,and it achieve good tracking in various scenarios.Finally,in order to further explore the tracking algorithm of the dynamic vision sensor in dark and reflective special scenes,my work studies the optical flow detection and feature tracking algorithm of the dynamic vision sensor.This algorithm only takes the event stream as input,and realizes feature tracking through optical flow detection.In order to realize optical flow detection,a probabilistic model is proposed in this chapter.The probability model is used to establish data association between events,and then iterative calculation is used to detect optical flow.In order to improve the tracking effect,we optimizes the algorithm by iterative closest point principle,and finally achieves a good tracking results.The tracking results is good in various scenarios,including the dark and reflective environment.