Matched Filters for Direct-Detection LiDAR in Non-Perpendicular Measurement Scenes

Abstract: The ongoing development of Advanced Driver Assistance Systems (ADAS) and autonomous vehicles pushes the requirements of accurate measurements from several on-board sensors, one of which being the Light Detection And Ranging (LiDAR). The performance of the LiDAR measurements relies partially on the reflectivity of the target, the level of background noise and the angle of incidence to the measurement scene. This thesis investigates if the temporal pulse shape due to non-perpendicular measurement scenes can be used to construct matched filters and how these filters affects the performance of the direct-detection LiDAR measurements. The temporal pulse shape was obtained by creating a simulation environment which replicates different measurement scenes by simulating the propagation of a laser pulse. The geometry of the measurement scene was derived from the radiometric concept of foreshortened area and the stochastic nature of the laser pulse was implemented using a Poisson process. Different matched filters were constructed by averaging the detected temporal pulse shape for different angles. The validity of the simulated temporal pulse shape was evaluated against real LiDAR measurements. The performance of the filters was evaluated both in terms of signal-to-noise ratio (SNR) and the receiver operating characteristics (ROC). It was found that the general behavior of the temporal pulse shape were similar for the simulated and the real LiDAR data. Some improvements and a more accurate model of the detector is probably necessary in order to improve the accuracy of the simulation environment. It was also found that all of the created matched filters increased both the SNR and the ROC in relation to not filtering at all. Further, it was found that for angles > 87° the correctly matched filter gave the best improvement.