Recently, the advancement of deep learning in discriminative feature learning from 3D LiDAR data has led to rapid development in the field of autonomous driving. However, automated processing uneven, unstructured, noisy, and massive 3D point clouds is a challenging and tedious task. In this paper, we provide a systematic review of existing compelling deep learning architectures applied in LiDAR point clouds, detailing for specific tasks in autonomous driving such as segmentation, detection, and classification. Although several published research papers focus on specific topics in computer vision for autonomous vehicles, to date, no general survey on deep learning applied in LiDAR point clouds for autonomous vehicles exists. Thus, the goal of this paper is to narrow the gap in this topic. More than 140 key contributions in the recent five years are summarized in this survey, including the milestone 3D deep architectures, the remarkable deep learning applications in 3D semantic segmentation, object detection, and classification; specific datasets, evaluation metrics, and the state of the art performance. Finally, we conclude the remaining challenges and future researches.

No abstract

With the rapid development of intelligent vehicles, drivers are increasingly likely to share their control authorities with the intelligent control unit. For building an efficient Advanced Driver Assistance Systems (ADAS) and shared-control systems, the vehicle needs to understand the drivers' intent and their activities to generate assistant and collaborative control strategies. In this study, a driver intention inference system that focuses on the highway lane change maneuvers is proposed. First, a high-level driver intention mechanism and framework are introduced. Then, a vision-based intention inference system is proposed, which captures the multi-modal signals based on multiple low-cost cameras and the VBOX vehicle data acquisition system. A novel ensemble bi-directional recurrent neural network (RNN) model with Long Short-Term Memory (LSTM) units is proposed to deal with the time-series driving sequence and the temporal behavioral patterns. Naturalistic highway driving data that consists of lane-keeping, left and right lane change maneuvers are collected and used for model construction and evaluation. Furthermore, the driver's pre-maneuver activities are statistically analyzed. It is found that for situation-aware, drivers usually check the mirrors for more than six seconds before they initiate the lane change maneuver, and the time interval between steering the handwheel and crossing the lane is about 2 s on average. Finally, hypothesis testing is conducted to show the significant improvement of the proposed algorithm over existing ones. With five-fold cross-validation, the EBiLSTM model achieves an average accuracy of 96.1% for the intention that is inferred 0.5 s before the maneuver starts.

No abstract

No abstract