Depth Map Reconstruction Based on a Computational Model of a Chaotic Laser Ranging System
Jiang Bowen,Yue Tao,Hu Xuemei
DOI: https://doi.org/10.3788/gzxb20225106.0628001
IF: 0.6
2022-01-01
ACTA PHOTONICA SINICA
Abstract:Radar is a sensor that uses electromagnetic waves for detection and ranging. The Light Radar (LIDAR) has been widely applied in many fields, such as robotics, ocean detection, atmospheric detection, intelligent driving, etc. Recently, LIDAR, based on the aperiodic random signal, has aroused great attention. The chaotic signal is one of the various aperiodic random signals, and the LIDAR systems taking the chaotic signal as the detection signal are named chaotic laser ranging systems. Considerable simulation and experimental results have illustrated that this kind of LIDAR system can perform attractive qualities, such as anti-jamming properties, high precision (mm-level), and multi-target real-time ranging ability. Nevertheless, up to now, existing work has not proposed a simulation model based on a realistic physical processes for chaotic laser ranging systems yet; also, there is no work that quantitatively analyzes the main degradation factors affecting the accuracy of chaotic laser ranging systems and the quality of reconstructed depth maps. In order to solve the problems above, a computational model based on the physical process for chaotic laser ranging systems is proposed in this paper. The computational model comprehensively considers various factors which will possibly cause chaotic signal degradation and ranging error during the realistic ranging process, including atmospheric attenuation, atmospheric turbulence, geometric attenuation, surface information of the object and its Bidirectional Reflection Distribution Function (BRDF), multipath noise, ambient noise, thermal noise and the degradation model of photodiodes. The program of the computational model is implemented with MATLAB. Among various degradation factors, there are three factors that require special awareness which namely BRDF, ambient noise, and multipath noise. In order to explore the influence of these three degradation factors on the accuracy of depth map reconstruction, this paper further uses the discrete chaotic sequence generated from the simulated Chua's chaotic circuit as the detection signal to scan the synthesized depth images and reconstructs depth maps by means of the cross-correlation mathematical method. To comprehensively assess the quality of the depth maps reconstructed under different degradation factors and degradation levels, we not only calculate the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM) between the reconstruction result and ground truth but also visualize both the reconstructed and ground-truth depth images. The experimental results show that the quality of depth maps reconstructed by the chaotic laser simulation ranging system is slightly affected by the roughness coefficient in BRDF model, but as the roughness coefficient increases, the influence of multipath noise will become non-negligible. Additionally, the chaotic laser simulation ranging system studied in this paper performs satisfying robustness against ambient noise when it is not extremely intense. However, the chaotic ranging system is relatively sensitive to multipath noise, i.e., even when the multipath noise is not intense, the depth map reconstruction quality will decrease rapidly. Therefore, when designing a realistic chaotic laser ranging system in practice, it is necessary to take the reflection and geometric property of the object and the influence of multipath noise into careful consideration. In conclusion, the computational model of a chaotic laser ranging system proposed and analyzed in this paper can serve as an important reference for analyzing the degradation factors affecting the ranging quality before designing and implementing a practical chaotic laser ranging system. Moreover, with the help of this computational model, it is possible for researchers to quickly and efficiently generate synthetic chaotic laser ranging datasets similar to the data measured in a realistic environment.