Research On Scene Reconstruction And High-speed Target Tracking Method For Pulse Image Sensor

As an important branch of machine vision,high-speed target tracking is a key technology to realize intelligent monitoring and image analysis.As the important part of the target tracking system,the performance of the image sensor will directly affect the result of target tracking.Among them,the pulse image sensor has a higher frame rate and lower data throughput than the traditional image sensors,so it has an advantage that traditional image sensors do not have in the field of high-speed target tracking.However,existing target tracking algorithms cannot be directly applied to the pulse image sensor,mainly because its single-bit pulse data does not directly contain grayscale information,which requires image reconstruction,but the existing image reconstruction methods will make the image surface exist a lot of noise and image lag,which will reduce the image quality and also affect the accuracy and stability of target tracking.Therefore,in order to solve these problems,this thesis proposes a scene reconstruction and high-speed target tracking method.First,the principle of the pulse image sensor is studied.This thesis analyzes the noise and image lag in the image reconstruction process based on the imaging principle.Second,the traditional pulse image reconstruction algorithm is improved to reconstruct the image of the tracking scene.The thesis divides the scene into static background and dynamic foreground according to the pulse interval.For the reconstruction of the static background,this thesis corrects the inconsistency between pixels and removes the fixed pattern noise.For the reconstruction of the dynamic foreground,this thesis corrects the error interval and lag interval in the pulse interval sequence,and the bilateral filter is used to remove the noise.Finally,the traditional Mean Shift(MS)tracking algorithm is improved to track the target in the scene.The thesis combines the characteristics of the sensor pulse data and the visual background extractor algorithm with the MS tracking algorithm,and identifies and marks the target in each reconstructed scene image.The effectiveness of the method in this thesis is proved by experiments.In the image reconstruction experiment,the Standard deviation(STD)is used to measure the imaging quality,and it is compared with other image reconstruction methods.In the static 1107 Lux,704 Lux,and 100 Lux uniform light reconstruction experiments,the STD of the image was reduced from 4.1367 to 0.5212,from 3.2312 to 0.5036,and from0.8644 to 0.5149,respectively.In the high-speed turntable reconstruction experiments at a rotation speed of 300r/min,500r/min,and 1000r/min,the STD of the image lag area was reduced from 10.0398 to 2.6040,from 13.6995 to 2.6968,and from 15.4981 to 2.7616,respectively.In the high-speed target tracking experiment,the Location standard deviation(LSTD)is used to measure the tracking accuracy and tracking stability.Compared with the traditional MS algorithm directly applied to the image sequence,the LSTD of the target tracking in the three scenes was reduced from 7.9897 to 2.0393,from 12.0790 to 3.7454,and from 14.4591 to 3.5654,respectively.In summary,the method in this thesis can effectively improve the imaging quality of the scene,tracking accuracy and stability of the target.