Refine Your Search

Topic

Author

Affiliation

Search Results

Journal Article

Real-time Pedestrian Detection using Convolutional Neural Network on Embedded Platform

2016-09-14
2016-01-1877
The convolutional neural network (CNN) has achieved extraordinary performance in image classification. However, the implementation of such architecture on embedded platforms is a big challenge task due to the computing resource constraint issue. This paper concentrates on optimization of CNN on embedded platforms with a case study of pedestrian detection in ADAS. The main contribution of this proposed CNN is its ability to run pedestrian classification task in real time with high accuracy based on a platform with ARM embedded. The CNN model has been trained with GPU locally and then transformed into an efficient implementation on embedded platforms. The efficient implementation uses dramatically small network scale and a lightweight CNN is obtained. Specifically, parameters of the network are compressed by adopting integer weights to reduce computational complexity. Meanwhile, other optimizations have also been proposed to adapt the general ARM processor architecture.
Journal Article

Active Noise Equalization of Vehicle Low Frequency Interior Distraction Level and its Optimization

2016-04-05
2016-01-1303
On the study of reducing the disturbance on driver’s attention induced by low frequency vehicle interior stationary noise, a subjective evaluation is firstly carried out by means of rank rating method which introduces Distraction Level (DL) as evaluation index. A visual-finger response test is developed to help evaluating members better recognize the Distraction Level during the evaluation. A non-linear back propagation artificial neural network (BPANN) is then modeled for the prediction of subjective Distraction Level, in which linear sound pressure RMS amplitudes of five Critical Band Rates (CBRs) from 20 to 500Hz are selected as inputs of the model. These inputs comprise an input vector of BPANN. Furthermore, active noise equalization (ANE) on DL is realized based on Filtered-x Least Mean Square (FxLMS) algorithm that controls the gain coefficients of inputs of trained BPANN.
Technical Paper

Study of Neural Network Control Algorithm in the Diesel Engine

2016-09-27
2016-01-8086
Based on BP neural network theory, a BP-PID control algorithm with strong self-learning and self-adapting ability is designed for the diesel engine speed governor. Nonlinear continuous functions can be approached with high precision by using this algorithm. The parameters of speed loop controller can be calibrated in real time through the BPPID algorithm. In order to verify the advantages of BP-PID control algorithm in reducing overshoot, increasing diesel engine dynamic characteristics and resisting disturbance, simulation model is built and experiments are carried out under initial condition, steady condition and condition with sudden load change. We compare the simulation results and the experiment results, and find they match each other. The results indicate that the transient speed regulation of the diesel engine can meet the requirements of stage power station by using BP-PID control algorithm.
Technical Paper

Modeling Thermal Engine Behavior Using Artificial Neural Network

2017-03-28
2017-01-0533
The knowledge of thermal behavior of combustion engines is extremely important e.g. to predict engine warm up or to calculate engine friction and finally to optimize fuel consumption. Typically, thermal engine behavior is modeled using look-up tables or semi-physical models to calculate the temperatures of structure, coolant and oil. Using look-up tables can result in inaccurate results due to interpolation and extrapolation; semi-physical modeling leads to high computation time. This work introduces a new kind of model to calculate thermal behavior of combustion engines using an artificial neural network (ANN) which is highly accurate and extremely fast. The neural network is a multi-layered feed-forward network; it is trained by data generated with a validated semi-physical model. Output data of the ANN-based model are calculated with nonlinear transformation of input data and weighting of these transformations.
Technical Paper

Predictive Analytics in Automobile Industry: A Comparison between Artificial Intelligence and Econometrics

2017-03-28
2017-01-0237
This study compares the model efficacy of Neural Network and Vector Auto Regression. Further it also analyses the impact of predictors controlling for total industry volume. Understanding both the methodologies has their distinctive advantages and disadvantages. Our empirical findings indicate that based on the characteristics of data such as non-stationary, non-linearity and non-normality paves the way for use of machine learning algorithm relative to econometrics technique. Our results suggest that data type and its characteristics are more important in determining the methodology than the methodology itself. In industry, econometrics methodologies are widely used due to their usage simplicity and its ability to explain the relationships in simple terms.
Technical Paper

Customer Data Driven PHEV Refuel Distance Modeling and Estimation

2017-03-28
2017-01-0235
Plug-in hybrid electric vehicles (PHEV) have an EV mode driving range which can cover a portion of customer daily driving. This EV mode range affects the refuel frequency substantially compared with conventional vehicle. For a conventional vehicle, daily driving pattern, tank size and fuel economy are the factors affecting the refuel frequency. While for a PHEV, EV range is another factor would affect the results substantially. Traditional method of label range can’t represent real world driving range between fill-ups for PHEV well. How to accurately predict the PHEV refuel distance taking into account real world customer driving patterns and PHEV parameters become critical for PHEV system design and optimization. This paper presents real world big customer data based PHEV refuel distance estimation modeling. The target is to estimate PHEV refuel distance given several specific parameters such as EV range, hybrid mode fuel economy, tank size etc.
Technical Paper

Estimation of Excavator Manipulator Position Using Neural Network-Based Vision System

2016-09-27
2016-01-8121
A neural network-based computer vision system is developed to estimate position of an excavator manipulator in real time. A camera is used to capture images of a manipulator, and the images are down-sampled and used to train a neural network. Then, the trained neural network can estimate the position of the excavator manipulator in real time. To study the feasibility of the proposed system, a webcam is used to capture images of an excavator simulation model and the captured images are used to train a neural network. The simulation results show that the developed neural network-based computer vision system can estimate the position of the excavator manipulator with an acceptable accuracy.
Technical Paper

Vision Based Object Distance Estimation

2017-03-28
2017-01-0108
This work describes a single camera based object distance estimation system. As technology on vehicles is constantly advancing on the road to autonomy, it is critical to know the locations of objects in 3D space for safe behavior of the vehicle. Though significant progress has been made on object detection in 2D sensor space from a single camera, this work additionally estimates the distance to said object without requiring stereo vision or absolute knowledge of vehicle motion. Specifically, our proposed system is comprised of three modules: vision based ego-motion estimation, object-detection, and distance estimation. In particular, we compensate for the vehicle ego-motion by using pin-hole camera model to increase the accuracy of the object distance estimation.
Technical Paper

A Bootstrap Approach to Training DNNs for the Automotive Theater

2017-03-28
2017-01-0099
The proposed technique is a tailored deep neural network (DNN) training approach which uses an iterative process to support the learning of DNNs by targeting their specific misclassification and missed detections. The process begins with a DNN that is trained on freely available annotated image data, which we will refer to as the Base model, where a subset of the categories for the classifier are related to the automotive theater. A small set of video capture files taken from drives with test vehicles are selected, (based on the diversity of scenes, frequency of vehicles, incidental lighting, etc.), and the Base model is used to detect/classify images within the video files. A software application developed specifically for this work then allows for the capture of frames from the video set where the DNN has made misclassifications. The corresponding annotation files for these images are subsequently corrected to eliminate mislabels.
Technical Paper

Region Proposal Technique for Traffic Light Detection Supplemented by Deep Learning and Virtual Data

2017-03-28
2017-01-0104
In this work, we outline a process for traffic light detection in the context of autonomous vehicles and driver assistance technology features. For our approach, we leverage the automatic annotations from virtually generated data of road scenes. Using the automatically generated bounding boxes around the illuminated traffic lights themselves, we trained an 8-layer deep neural network, without pre-training, for classification of traffic light signals (green, amber, red). After training on virtual data, we tested the network on real world data collected from a forward facing camera on a vehicle. Our new region proposal technique uses color space conversion and contour extraction to identify candidate regions to feed to the deep neural network classifier. Depending on time of day, we convert our RGB images in order to more accurately extract the appropriate regions of interest and filter them based on color, shape and size. These candidate regions are fed to a deep neural network.
Technical Paper

Augmentation of an Artificial Neural Network (ANN) Model with Expert Knowledge of Critical Combustion Features for Optimizing a Compression Ignition Engine Using Multiple Injections

2017-03-28
2017-01-0701
The objective of this work was to identify methods of reliably predicting optimum operating conditions in an experimental compression ignition engine using multiple injections. Abstract modeling offered an efficient way to predict large volumes data, when compared with simulation, although the initial cost of constructing such models can be large. This work aims to reduce that initial cost by adding knowledge about the favorable network structures and training rules which are discovered. The data were gathered from a high pressure common rail direct injection turbocharged compression ignition engine utilizing a high EGR configuration. The range of design parameters were relatively large; 100 MPa - 240 MPa for fuel pressure, up to 62% EGR using a modified, long-route, low pressure EGR system, while the pilot timing, main timing, and pilot ratio were free within the safe operating window for the engine.
Technical Paper

Investigation of Vehicle Speed Prediction from Neural Network Fit of Real World Driving Data for Improved Engine On/Off Control of the EcoCAR3 Hybrid Camaro

2017-03-28
2017-01-1262
The EcoCAR3 competition challenges student teams to redesign a 2016 Chevrolet Camaro to reduce environmental impacts and increase energy efficiency while maintaining performance and safety that consumers expect from a Camaro. Energy management of the new hybrid powertrain is an integral component of the overall efficiency of the car and is a prime focus of Colorado State University’s (CSU) Vehicle Innovation Team. Previous research has shown that error-less predictions about future driving characteristics can be used to more efficiently manage hybrid powertrains. In this study, a novel, real-world implementable energy management strategy is investigated for use in the EcoCAR3 Hybrid Camaro. This strategy uses a Nonlinear Autoregressive Artificial Neural Network with Exogenous inputs (NARX Artificial Neural Network) trained with real-world driving data from a selected drive cycle to predict future vehicle speeds along that drive cycle.
Technical Paper

Emissions Modeling of a Light-Duty Diesel Engine for Model-Based Control Design Using Multi-Layer Perceptron Neural Networks

2017-03-28
2017-01-0601
The development of advanced model-based engine control strategies, such as economic model predictive control (eMPC) for diesel engine fuel economy and emission optimization, requires accurate and low-complexity models for controller design validation. This paper presents the NOx and smoke emissions modeling of a light duty diesel engine equipped with a variable geometry turbocharger (VGT) and a high pressure exhaust gas recirculation (EGR) system. Such emission models can be integrated with an existing air path model into a complete engine mean value model (MVM), which can predict engine behavior at different operating conditions for controller design and validation before physical engine tests. The NOx and smoke emission models adopt an artificial neural network (ANN) approach with Multi-Layer Perceptron (MLP) architectures. The networks are trained and validated using experimental data collected from engine bench tests.
Technical Paper

Free-Positioned Smartphone Sensing for Vehicle Dynamics Estimation

2017-03-28
2017-01-0072
With the embedded sensors – typically Inertial Measurement Units (IMU) and GPS, the smartphone could be leveraged as a low-cost sensing platform for estimating vehicle dynamics. However, the orientation and relative movement of the smartphone inside the vehicle yields the main challenge for platform deployment. This study proposes a solution of converting the smartphone-referenced IMU readings into vehicle-referenced accelerations, which allows free-positioned smartphone for the in-vehicle dynamics sensing. The approach is consisted of (i) geometry coordinate transformation techniques, (ii) neural networks regression of IMU from GPS, and (iii) adaptive filtering processes. Experiment is conducted in three driving environments which cover high occurrence of vehicle dynamic movements in lateral, longitudinal, and vertical directions. The processing effectiveness at five typical positions (three fixed and two flexible) are examined.
Technical Paper

Distance Map Estimation of Stereoscopic Images Using Deep Neural Networks for Autonomous Vehicle Driving

2017-03-28
2017-01-0071
While operating a vehicle in either autonomous or occupant piloted mode, an array of sensors can be used to guide the vehicle including stereo cameras. The state-of-the-art distance map estimation algorithms, e.g. stereo matching, usually detect corresponding features in stereo images, and estimate disparities to compute the distance map in a scene. However, depending on the image size, content and quality, the feature extraction process can become inaccurate, unstable and slow. In contrast, we employ deep convolutional neural networks, and propose two architectures to estimate distance maps from stereo images. The first architecture is a simple and generic network that identifies which features to extract, and how to combine them in a multi-resolution framework. The second architecture is a more specialized one that extracts local similarity information from two images, which are used for stereo feature matching, and fuses them at multiple resolutions to generate the distance map.
Technical Paper

Development of an Autonomous Vehicle Control Strategy Using a Single Camera and Deep Neural Networks

2018-04-03
2018-01-0035
Autonomous vehicle development has benefited from sanctioned competitions dating back to the original 2004 DARPA Grand Challenge. Since these competitions, fully autonomous vehicles have become much closer to significant real-world use with the majority of research focused on reliability, safety and cost reduction. Our research details the recent challenges experienced at the 2017 Self Racing Cars event where a team of international Udacity students worked together over a 6 week period, from team selection to race day. The team’s goal was to provide real-time vehicle control of steering, braking, and throttle through an end-to-end deep neural network. Multiple architectures were tested and used including convolutional neural networks (CNN) and recurrent neural networks (RNN). We began our work by modifying a Udacity driving simulator to collect data and develop training models which we implemented and trained on a laptop GPU.
Journal Article

Adaptive Network Trained Controller for Automotive Steering Systems

2017-04-11
2017-01-9626
Electrical Power Assist Steering (EPAS) systems are currently eliminating the traditional hydraulic steering systems in vehicles. EPAS systems are nonlinear Multi Input Multi Output (MIMO) systems with multiple objectives, including fast response to the driver torque command, good driver feel, and attenuation of load disturbance and sensor noises. Optimal control method is employed to design EPAS system controllers for improved performance and robustness. But these controllers have showed acceptable performance for certain operating conditions and undesired steering feel for high steering gain. In this work, the neural networks are used which replace the optimal controllers of EPAS systems. A Euclidean adaptive resonance theory (EART) networks is trained according to the data collected from an H∞ optimal controller. The collected data represent the controller input and output signals. The said data are normalized and clustered into categories in the EART modules.
Technical Paper

Economic and Efficient Hybrid Vehicle Fuel Economy and Emissions Modeling Using an Artificial Neural Network

2018-04-03
2018-01-0314
High accuracy hybrid vehicle fuel consumption (FC) and emissions models used in practice today are the product of years of research, are physics based, and bear a large computational cost. However, it may be possible to replace these models with a non-physics based, higher accuracy, and computationally efficient versions. In this research, an alternative method is developed by training and testing a time series artificial neural network (ANN) using real world, on-road data for a hydraulic hybrid truck to predict instantaneous FC and emissions. Parameters affecting model fidelity were investigated including the number of neurons in the hidden layer, specific training inputs, dataset length, and hybrid system status. The results show that the ANN model was computationally faster and predicted FC within a mean absolute error of 0-0.1%. For emissions prediction the ANN model had a mean absolute error of 0-3% across CO2, CO, and NOx aggregate predicted concentrations.
Technical Paper

Enabling Prediction for Optimal Fuel Economy Vehicle Control

2018-04-03
2018-01-1012
Vehicle control using prediction based optimal energy management has been demonstrated to achieve better fuel economy resulting in economic, environmental, and societal benefits. However, research focusing on prediction derivation for use in optimal energy management is limited despite the existence of hundreds of optimal energy management research papers published in the last decade. In this work, multiple data sources are used as inputs to derive a prediction for use in optimal energy management. Data sources include previous drive cycle information, current vehicle state, the global positioning system, travel time data, and an advanced driver assistance system (ADAS) that can identify vehicles, signs, and traffic lights. To derive the prediction, the data inputs are used in a nonlinear autoregressive artificial neural network with external inputs (NARX).
Journal Article

Machine Learning for Misfire Detection in a Dynamic Skip Fire Engine

2018-04-03
2018-01-1158
Dynamic skip fire (DSF) has shown significant fuel economy improvements via reduction of pumping losses that generally affect throttled spark-ignition engines. For production readiness, DSF engines must meet regulations for on-board diagnostics (OBD-II), which require detection and monitoring of misfire in all passenger vehicles powered by an internal combustion engine. Numerous misfire detection methods found in the literature, such as those using peak crankshaft angular acceleration, are generally not suitable for DSF engines due to added complexity of skipping cylinders. Specifically, crankshaft acceleration traces may change abruptly as the firing sequence changes. This article presents a novel method for misfire detection in a DSF engine using machine learning and artificial neural networks. Two machine learning approaches are presented.
X