Technical Paper collections have been re-named for better clarity and alignment.x

Your Selections

Neural networks
Show Only

Collections

File Formats

Content Types

Dates

Sectors

Topics

Authors

Publishers

Affiliations

Committees

Events

Magazine

   This content is not included in your SAE MOBILUS subscription, or you are not logged in.

Sparse-CNN based Real-time Semantic Segmentation of Point Cloud from LiDAR

Hanyang university-Jaehyun Park, Chansoo Kim, Myoungho Sunwoo
Konkuk Univ-Kichun Jo
  • Technical Paper
  • 2020-01-0725
To be published on 2020-04-14 by SAE International in United States
A Light Detection And Ranging (LiDAR) sensor, which provides the precise location and shape information of objects near the ego-vehicle based on a point cloud, plays a core sensor in the perception of the autonomous driving. However, the point cloud cannot provide the semantic information of objects such as buildings, vehicles, and pedestrians. The absence of the semantic information can cause the degraded performances of object detection and motion prediction. For the extraction of the semantic information from the point cloud, which is called semantic segmentation, many deep learning-based methods have been researched. Since most of the deep learning-based methods have deep and bulky networks, the methods may cause insufficient real-time performance. Moreover, while the deep learning-based methods need large amounts of the labeled point cloud to train their models, the previous methods trained with a limited amount of the labeled point clouds which are obtained from a real driving environment. This shortage of the labeled point cloud may cause difficulty in the improvement of the performance. In our research, we propose the sparse-CNN based…
   This content is not included in your SAE MOBILUS subscription, or you are not logged in.

Vehicle Velocity Prediction Using Artificial Neural Network and Effect of Real World Signals on Prediction Window.

CEAS Western Michigan University-Alvis Fong
Colorado State Univ-Thomas Bradley
  • Technical Paper
  • 2020-01-0729
To be published on 2020-04-14 by SAE International in United States
Prediction of vehicle velocity is important since it can realize improvements in the fuel economy/energy efficiency, drivability and safety. Velocity prediction has been addressed in many publications. Several references considered deterministic and stochastic approaches such as Markov chain, autoregressive models, and artificial neural networks. There are numerous new sensor and signal technologies like vehicle-to-vehicle and vehicle-to-infrastructure communication that can be used to obtain inclusive datasets. Using these inclusive datasets of sensors in deep neural networks, high accuracy velocity predictions can be achieved. This research builds upon previous findings that Long Short-Term Memory (LSTM) deep neural networks provide the highest velocity prediction fidelity. We developed LSTM deep neural network which uses different groups of datasets collected in Fort Collins. Synchronous data was gathered using a test vehicle equipped with sensors to measure ego vehicle position and velocity, ADAS-derived near-neighbor relative position and velocity, and infrastructure-level transit time and signal phase and timing. Effect of different group of datasets on forward velocity prediction window of 10, 15, 20 and 30 seconds is studied. Developed algorithm is tested…
   This content is not included in your SAE MOBILUS subscription, or you are not logged in.

Development of a Camera-based Driver State Monitoring System for Cost-Effective Embedded Solution

Hitachi America, Ltd.-Xunfei Zhou, Tobias Wingert, Maximilian Sauer, Subrata Kundu
  • Technical Paper
  • 2020-01-1210
To be published on 2020-04-14 by SAE International in United States
To prevent the severe consequences of unsafe driving behaviors, it is crucial to monitor and analyze the state of the driver. Developing an effective driver state monitoring (DSM) systems is particularly challenging due to limited computation capabilities of embedded systems in automobiles and the need for finishing processing in real-time. However, most of the existing research work was conducted in a lab environment with expensive equipment while lacking in-car benchmarking and validation. In this paper, a DSM system that estimates driver's alertness and drowsiness level as well as performs emotion detection built with a cost-effective embedded system is presented. The proposed system consists of a mono camera that captures driver's facial image in real-time and a machine learning based detection algorithm that detects facial landmark points and use that information to infer driver's state. In the detection module, driver's distraction level is evaluated by estimating head-pose through solving a perspective-n-point problem, drowsiness level is estimated by processing eyelid related parameters extracted from facial keypoints data, and two approaches were investigated for emotion recognition with performance…
   This content is not included in your SAE MOBILUS subscription, or you are not logged in.

Decision Making and Trajectory Planning of Intelligent Vehicle’ s Lane-changing Behavior on Highways under Multi-objective Constrains

Wuhan University of Technology-Linzhen Nie, Zhishuai Yin, Haoran Huang
  • Technical Paper
  • 2020-01-0124
To be published on 2020-04-14 by SAE International in United States
Discretionary lane-changing is commonly seen in highway driving. Intelligent vehicles are expected to change lanes discretionarily for better driving experience and higher traffic efficiency. This study proposes to optimize the decision making and trajectory planning process so that intelligent vehicles make lane changes not only with driving safety taken into account, but also with the goal to improve driving comfort as well as to meet the driver’ s expectation. The mechanism of how various factors contribute to the driver’s intention to change lanes is studied by carrying out a series of driving simulation experiments, and a Lane-changing Intention Generation (LIG) model based on Convolutional Neural Network (CNN) is proposed. The inputs of the CNN are data fragments of several influence factors including the relative speed and the distance between the subject vehicle and the preceding vehicles in current lane and both sides of the lane, and the type of the preceding vehicles in current lane and both sides of the lane, the average speed of the left and right traffic flow, the in a certain…
   This content is not included in your SAE MOBILUS subscription, or you are not logged in.

Driver Distraction Detection with a Two-stream Convolutional Neural Network

Wuhan University of Technology-Yuefeng Ma, Zhishuai Yin, Linzhen Nie
  • Technical Paper
  • 2020-01-1039
To be published on 2020-04-14 by SAE International in United States
Driver distraction detection is crucial to driving safety when autonomous vehicles are co-piloted. Recognizing drivers’ behaviors that are highly related with distraction from real-time video stream is widely acknowledged as an effective approach mainly due to its non-intrusiveness. In recently years, deep learning neural networks have been adopted to by-pass the procedure of designing features artificially, which used to be the major downside of computer-vision based approaches. However, the detection accuracy and generalization ability is still not satisfying since most deep learning models extracts only spatial information contained in images. This research develops a driver distraction model based on a two-stream, spatial and temporal, convolutional neural network (CNN). The CNN in both stream is improved with Batch Normalization-Inception (BN-Inception) modules which increase the sparsity in the inception modules in GoogLeNet, so that the network is further speeded up and also more adapted to features at various-scales. The original RGB image is fed into the spatial stream CNN to extract static information, and the feature map of optical flow field extracted from adjacent image frames is…
   This content is not included in your SAE MOBILUS subscription, or you are not logged in.

A Pre-warning Method for Cornering Speed of Concrete Mixer Truck

Suizhou-WUT Industry Research Institute-Gangfeng Tan
Wuhan University of Technology-Yifeng Jiang, Haoyu Wang, Zelong Wang, Zhenyu Wang, Ming Li
  • Technical Paper
  • 2020-01-1003
To be published on 2020-04-14 by SAE International in United States
The high gravity center of the concrete mixer truck reduces the truck’s stability while steering, and the rolling stirring tank makes the stability even worse than the regular engineering vehicle due to the dynamic variation of centroid position. Most of the researches on the rollover stability of concrete mixer trucks focus on the rollover model establishment and dynamics simulation module. The influence of concrete centroid changes is ignored when the safe cornering speed is calculated. This paper proposes a pre-warning method for cornering speed of concrete mixer truck based on centroid dynamic simulation. In the method, the mixing tank stirring model and the vehicle driving dynamics model are established on the Fluent and TruckSim simulation platforms, respectively.The theoretical speed threshold obtained by simulation is used as the evaluation index of warning of the speed for steering. First, the dynamic simulation of the stirring tank model is carried out by Fluent. According to Newton Leibniz numerical calculation method, Matlab is used to obtain the mathematical model of the centroid position and the main parameters of the…
   This content is not included in your SAE MOBILUS subscription, or you are not logged in.

Comparison of CNN and LSTM for modeling virtual sensors in an engine

Chalmers Univ. of Technology-Mauro Bellone, Yiannis Karayiannidis
Volvo AB Volvo Penta-Ethan Faghani
  • Technical Paper
  • 2020-01-0735
To be published on 2020-04-14 by SAE International in United States
Automotive industry makes extensive use of virtual models to increase the efficiency during the development stage. The complexity of such virtual models increases as the complexity of the process that they describe, and for this reason new methods for their development are constantly evaluated. Among many others, data-driven techniques and machine learning offer promising results, creating deep neural networks that map input-output relations. This works aims at evaluating the performance of different neural network architectures for the estimation of engine status and gas emissions. More specifically, Convolutional Neural Network (CNN) and Long-Short Term Memory (LSTM) will be evaluated in terms of performance, using different techniques to increase the model generalization. During the learning stage data from different engine cycles are fed to the neural network. To evaluate model generalization the network is then tested over new, previously unseen, engine cycles. Results show that our model over-performs a state of the art models, the best performance was found from the LSTM model with 2.40%, 2.80% and 18.19% error for flow fuel, NOx and soot sensor respectively.
   This content is not included in your SAE MOBILUS subscription, or you are not logged in.

Heterogeneous Machine Learning on High Performance Computing for End to End Driving of Autonomous Vehicles

National Renewable Energy Laboratory-Xiangyu Zhang, Peter Graf
Oak Ridge National Laboratory-Robert Patton, Shang Gao, Spencer Paulissen, Nicholas Haas, Brian Jewell
  • Technical Paper
  • 2020-01-0739
To be published on 2020-04-14 by SAE International in United States
Current artificial intelligence techniques for end to end driving of autonomous vehicles typically rely on a single form of learning or training processes along with a corresponding dataset or simulation environment. Relatively speaking, success has been shown for a variety of learning modalities in which it can be shown that the machine can successfully “drive” a vehicle. However, the realm of real-world driving extends significantly beyond the realm of limited test environments for machine training. This creates an enormous gap in capability between these two realms. With their superior neural network structures and learning capabilities, humans can be easily trained within a short period of time to proceed from limited test environments to real world driving. For machines though, this gap is guarded by at least two challenges: 1) machine learning techniques remain brittle and unable to generalize to a wide range of scenarios, and 2) effective training data that enhances generalization and generates the desired driving behavior. Further, each challenge can be computationally intensive on its own thereby exasperating the gap. Moreover, is has…
   This content is not included in your SAE MOBILUS subscription, or you are not logged in.

Driver Visual Focus of Attention Estimation in Autonomous Vehicles

Dura Automotive Systems-Iyad Mansour
Oakland University-Alaaldin Hijaz, WING-YUE LOUIE, Matthew Bellafaire, Osamah Rawashdeh
  • Technical Paper
  • 2020-01-1037
To be published on 2020-04-14 by SAE International in United States
Abstract-An existing challenge in current state-of-the-art autonomous vehicles is the process of safely transferring control from autonomous driving mode to manual mode because the driver may be distracted with secondary tasks. Such distractions may impair a driver’s situational awareness of the driving environment which will lead to fatal outcomes during a handover. Current state-of-the-art vehicles notify a user of an imminent handover via auditory, visual, and physical alerts but are unable to improve a driver’s situational awareness before a handover is executed. The overall goal of our research team is to address the challenge of providing a driver with relevant information to regain situational awareness of the driving task. In this paper, we introduce a novel approach to estimating a driver’s visual focus of attention using a 2D RGB camera as input to a Multi-Input Convolutional Neural Network with shared weights. The system was validated in a realistic driving scenario. The developed approach is a first step towards estimating a driver’s situational awareness from their observable indicators which will in the future be utilized to…
   This content is not included in your SAE MOBILUS subscription, or you are not logged in.

Kalman Filter Slope Measurement Method Based on Improved Genetic Algorithm- Back Propagation

Suizhou-WUT Industry Research Institute-Gangfeng Tan
Wuhan University of Technology-Haoyu Wang, Donghua Guo, Zhenyu Wang, Ming Li, Yifeng Jiang, Meng Ye, Kailang Chen
  • Technical Paper
  • 2020-01-0897
To be published on 2020-04-14 by SAE International in United States
How to improve the measurement accuracy of road grade is the key content of the research on the speed warning of commercial vehicles in mountainous roads. If there is a large measurement error, the obtained speed threshold will be biased, posing a safety hazard. Conventional measuring instruments such as accelerometers and gyroscopes generally have noise fluctuation interference or time accumulation error, resulting in large measurement errors. In response to this situation, the Kalman filter method is often used for filtering to reduce the interference of unwanted signals, thereby improving the accuracy of the slope measurement. However, the Kalman filtering method is limited by the estimation error of various parameters, and the filtering effect is difficult to meet the project research requirements. In this paper, the acceleration of vehicle gravity, running speed and acceleration of parallel slope are used as auxiliary measurement parameters to improve the measurement method of mountain road slope. Based on the Kalman model, genetic algorithm (GA) and BP neural network are used to carry out the innovation , covariance matrix and the…