Research On Non-line-of-sight Tracking Technology Based On Photon Measurement

The traditional imaging technology usually images the light emitted or scattered by the object in the line of sight,while the light emitted by the object hidden behind the wall can only enter the detector after scattering through the relay surface.The information of objects out of sight(non-line-of-sight)is of great value to some special environments,such as automatic driving and rescue.Knowing the information of non-line-of-sight environment in advance can increase decision-making time,and it is possible to deal with emergencies in advance,improve efficiency and even reduce casualties.Therefore,non-line-of-sight detection has always been the goal that researchers hope to achieve.With the improvement of time resolution and sensitivity of electro-optic detection equipment,it is possible to detect non-line-of-sight environment through the relay wall.Through the method of active detection,the researchers have realized the detection of weak beams on the relay wall with detectors of high sensitivity and high time resolution such as streak camera and single photon avalanche diode.They have successively proposed back projection algorithm,light cone transformation algorithm,Fermat flow algorithm,virtual wave vector field imaging and other methods to detect the non-line-of-sight environment.They have also proposed the ellipse positioning method to detect the position of the object based on the model of photon flight.The research content of this paper is locating the non-line-of-sight objects stably and efficiently.The main work of this paper is as follows:(1)Firstly,the research background and significance of non-line-of-sight detection are discussed.This paper summarizes and introduces the principle and implementation scheme of several current mainstream non-line-of-sight reconstruction algorithms,and several optimized algorithms based on the most commonly used back projection method are introduced and analyzed.(2)The basic theories of non-line-of-sight track are introduced,including ellipse location,illumination model,scattering model,etc.Various optical phenomena on the surface of the object are introduced systematically,which provides the choices for the simulation with the relay walls and the targets of different material properties.(3)The optical process of non-line-of-sight detection is optimized by the means of photometry and measured bidirectional reflectance distribution function.In order to increase the reality of simulation,the irradiated surface of non-line-of-sight target is divided into a number of small surface elements.The third scattering signal is simulated according to the parameters of real equipment and the measured bidirectional reflection distribution functions.The influence of different time-of-flight extraction methods on the positioning accuracy and stability,and the influence of environmental geometry parameters,equipment parameters,and algorithm parameters on positioning effect are researched according to the simulated signal.(4)The experimental platform of non-line-of-sight location in short distance is built.The transmittance of the lens designed for visible light is measured when it works in the infrared band.It solves the problem of calibration between the zero point of optical path in free space and the zero point of time-of-flight when the optical path in waveguide and the delay of experimental equipment are unknown.The feasibility of ellipse location method is verified,and the time-of-flight extraction method based on filtering algorithm is compared with the traditional one based on the Gauss fitting method.It is concluded that there is a contradiction between the unmanned degree of time-of-flight extraction and location stability when the traditional method is applied under the influence of unstable background and strong noise without prior information and human recognition.It shows the adaptability of the filtering method to the unstable background and the advantages in positioning accuracy and stability without the recognition of target interval and prior information.(5)The experimental platform of non-line-of-sight tracking in long distance is built.The location scheme is optimized in terms of signal coupling between each channel,low signal-to-noise ratio,large error in zero point calibration and lowefficiency of reading and writing data manually,when the strategy of location in short distance is applied to locate a hidden object at a long range.The method of calibrating the relative zero point according to the feature of background signal is proposed to solve the problems of signal coupling and zero calibration.The methods of mean filtering with big window to restrain data jitter,triangle discrimination to filter out high-intensity background signal and improve signal-to-noise ratio,and threshold segmentation to extract target signal accurately are proposed.The automatic real-time positioning software is developed to read and write multi-channel data automatically in real time,which improves the efficiency of positioning.The real-time location of hidden target in long distance is realized,which shows the advantage of filtering algorithm in automation.