The performance of LiDAR degrades severely in adverse weather conditions of which fog is considered one of the worst. In this letter, fog effects on Single Photon Avalanche Diode (SPAD) LiDAR are studied, and techniques to mitigate the effects caused by uniform fog are proposed. Fog mainly imposes two challenges to LiDAR in vehicular applications, namely, severe signal attenuation at long ranges and false alarms caused by fog at short ranges. Different strategies are needed to address these issues. A Generalized Likelihood Ratio Test (GLRT) with a Constant False Alarm Rate (CFAR) threshold is proposed for target detection at low signal levels. While for reducing false alarms at short ranges, a defogging technique based on fog profile estimation is developed. Experiment and simulation results demonstrate the potentials and limitations of these techniques at combating fog.
A method of direction-of-arrival (DoA) estimation for FMCW (Frequency Modulated Continuous Wave) radar is presented. In addition to MUSIC, which is the popular high-resolution DoA estimation algorithm, deep learning has recently emerged as a very promising alternative. It is proposed in this paper to use a 3D convolutional neural network (CNN) for DoA estimation. The 3D-CNN extracts from the radar data cube spectrum features of the region of interest (RoI) centered on the potential positions of the targets, thereby capturing the spectrum phase shift information, which corresponds to DoA, along the antenna axis. Finally, the results of simulations and experiments are provided to demonstrate the superior performance, as well as the limitations, of the proposed 3D-CNN.
High-performance light detection and ranging (LiDAR) modules are highly demanded for advanced driving assistance system and autonomous driving. With CMOS-fabricated single-photon avalanche diodes (SPADs) and pulsed-operated semiconductor lasers, photon-counting based vehicle LiDARs are one of the most promising candidates for low-cost solution. The high data rate and the long integration time of SPAD LiDARs are two of main obstacles for realizing commercially affordable 3-D imagers. In this work, we develop an evaluation method for the minimum ranging time and investigate the time-bin width effect on the ranging precision. A simple formula for predicting the shortest integration time is proposed and compared with the experimental data. We also demonstrate that a proper time-bin width can reduce the data rate and required memory size without degrading the distance resolution. Our work paves the way for making low-cost vehicle LiDARs through system optimizing and engineering.
The performance of LiDAR with single-photon avalanche diode (SPAD) detectors can degrade seriously when a shortage of photons occurs in adverse environments. It is desirable for SPAD LiDAR to accumulate enough photons so that reliable target detection is maintained under such conditions. It is proposed in this letter to adopt random laser-emitting schemes, such as stochastic pulse position modulation (SPPM), to have more laser pulses illuminated on targets to increase reflected photons. Accordingly, at the receiver side, a novel method, namely multi-start-multi-stop (MSMS), is proposed to work with SPPM to generate histograms for further processing of target detection and ranging. Theoretical analysis and experiments show that the SPPM+MSMS LiDAR can produce, with far less integration time, signal levels comparable to that of conventional LiDAR.
As autonomous driving technology progresses forward, frequency modulated continuous wave (FMCW) radar is projected to be used more widely for automotive purposes. Due to the expected rapid growth of road vehicles equipped with radars, more attention is paid to finding ways of reducing mutual interference among automotive radars. In this letter, a novel scheme is proposed to solve the issue of velocity estimation for FMCW radar with random pulse position modulation, which is a promising technique to drastically mitigate mutual interference. The proposed scheme uses a two-dimensional convolutional neural network working on covariance matrices of signals extracted from the region of interest as well as the information of chirp positions. Analysis of its performance, in particular comparison with that of orthogonal matching pursuit, with simulation and experimental data demonstrate the potential of the approach.