Research On Algorithm And Technique Of Non-Line-of-Sight Imaging

Traditional photoelectric imaging system,limited by the nature of the linear propagation of light,can only record the information of objects within the field of view.When the light path between the imaging target and the imaging system is blocked by obstacles,the traditional imaging system can not detect and image the target.In recent years,with the booming development of photoelectric detection technology and computational imaging technology,non-line-of-sight imaging technology has come into being.Unlike traditional imaging technologies that focus only on targets within the field of view of the imaging system,non-line-of-sight imaging technology can image and reconstruct hidden targets that are obscured by obstacles.This emerging imaging technology expands human imaging detection capabilities to a new dimension.The technology has a wide range of applications,such as autonomous driving,medical detection,disaster relief,etc.,and has therefore generated extensive research interest in the academic community.The basic conception of non-line-of-sight imaging was first presented at the International Conference on Computer Vision in 2009.In 2012,researchers achieved the first experimental demonstration of 3D non-line-of-sight imaging.Since then,researchers around the world have conducted in-depth and detailed research on non-lineof-sight imaging from both system and algorithm aspects.After more than ten years of development,the performance and stability of non-line-of-sight imaging system technology has been greatly improved.However,although some progress has been made in the study of non-line-of-sight imaging,there are still a series of scientific and technical problems that prevent the technology from being useful in practical applications.At first,most existing NLOS imaging systems can only detect hidden objects over a very short range.Secondly,the imaging speed of NLOS imaging is slow.Thirdly,most systems need a complicated pre-calibration process before detection.In this thesis,the authors and the team have made the following innovative work and studies to address the above scientific and technical issues.First,the authors designed a high-efficiency and low-noise optical detection system for the problem of short imaging distance,developed a long-range non-line-of-sight imaging algorithm and demonstrated long-range non-line-of-sight imaging experiment.This work extended the imaging distance of non-visual imaging.Second,the authors developed a single-photon non-line-of-sight imaging algorithm.This algorithm can reduce the photon numbers required by non-line-of-sight imaging system which can reduce the acquisition time of the non-line-of-sight imaging system.The authors also built an integrated non-line-of-sight imaging system and demonstrated single-photon non-line-of-sight imaging.Finally,to avoid the manual pre-calibration process,the authors proposed an autopre-calibration algorithm,which can automatically pre-calibrate before detection,thus reducing the tedious pre-calibration process.Based on the auto-calibration algorithm,the authors optimized the weight,size and power of the NLOS imaging system for unmanned aerial vehicle(UAV).