Search Results

As a key component of in-vehicle intelligent voice technology, speech enhancement can extract clean speech signals contaminated by environmental noise to improve the perceptual quality and intelligibility of speech. It has extensive applications in the field of intelligent car cabins. Although some end-to-end speech enhancement methods based on time domain have been proposed, there is often limited consideration given to designing model architectures based on the characteristics of the speech signal. In this paper, we propose a new U-Net based speech enhancement framework that utilizes the temporal correlation of speech signals to reconstruct higher-quality and more intelligible clean speech.

As one of the pollutants that cannot be ignored, soot has a great impact on human health, environment, and energy conversion. In this investigation, the effect of residence time (25ms, 35ms, and 45ms) and ammonia on morphology and nanostructure of soot in laminar ethylene flames has been studied under atmospheric conditions and different flame heights (15 mm and 30 mm). The transmission electron microscopy (TEM) and high-resolution transmission electron microscope (HRTEM) are used to obtain morphology of aggregates and nanostructure of primary particles, respectively. In addition, to analyze the nanostructure of the particles, an analysis program is built based on MATLAB software, which is able to obtain the fringe separation distance, fringe length, and fringe tortuosity parameters of primary particles, and has been verified by the multilayer graphene interlayer distance.

LiDAR and camera fusion have emerged as a promising approach for improving place recognition in robotics and autonomous vehicles. However, most existing approaches often treat sensors separately, overlooking the potential benefits of correlation between them. In this paper, we propose a Cross- Modality Module (CMM) to leverage the potential correlation of LiDAR and camera features for place recognition. Besides, to fully exploit potential of each modality, we propose a Local-Global Fusion Module to supplement global coarse-grained features with local fine-grained features. The experiment results on public datasets demonstrate that our approach effectively improves the average recall by 2.3%, reaching 98.7%, compared with simply stacking of LiDAR and camera.

Centrifugal Pendulum Vibration Absorber (CPVA for short) is used to absorb torsional vibrations caused by the shifting motion of the engine. It is increasingly used in modern powertrains. In the research of the dynamic characteristics of the CPVA, it is necessary to obtain the real motion of the pendulum to compensate the fitting performance of mathematical model. The usual method is to install an angle sensor to measure the movement of the pendulum. On the one hand, the installation of the sensor will affect its movement to a certain extent, so that the measurement results do not match the actual motion. On the other hand, the motion of the pendulum is not only the rotational motion around the rotational axis of the CPVA rotor, but also has translation relative to it. As a result, it is difficult to obtain accurate motion only by the angle sensor. We proposed a non-contact centrifugal pendulum motion measurement method.

High-speed vehicles in low illumination environments severely blur the images used in object detectors, which poses a potential threat to object detector-based advanced driver assistance systems (ADAS) and autonomous driving systems. Augmenting the training images for object detectors is an efficient way to mitigate the threat from motion blur. However, little attention has been paid to the motion of the vehicle and the position of objects in the traffic scene, which limits the consistence between the resulting augmented images and traffic scenes. In this paper, we present a vehicle kinematics-based image augmentation algorithm by modeling and analyzing the traffic scenes to generate more realistic augmented images and achieve higher robustness improvement on object detectors against motion blur. Firstly, we propose a traffic scene model considering vehicle motion and the relationship between the vehicle and the object in the traffic scene.

Image corruptions due to noise, blur, contrast change, etc., could lead to a significant performance decline of Deep Neural Networks (DNN), which poses a potential threat to DNN-based autonomous vehicles. Previous works attempted to explain corruption from a Fourier perspective. By comparing the absolute Fourier spectrum difference between corrupted images and clean images in the RGB color space, they regard the noise from some corruptions (Gaussian noise, defocus blur, etc.) as concentrating on the high-frequency components while others (contrast, fog, etc.) concentrate on the low-frequency components. In this work, we present a new perspective that unifies corruptions as noise from high frequency and thus propose an image augmentation algorithm to achieve a more robust performance against common corruptions. First, we notice the 1/fα statistical rule of the natural image's spectrum and the channels-wise differential sensitivity on the YCbCr color space of the Human Visual System.

Visual sensors are widely used in autonomous vehicles (AVs) for object detection due to the advantages of abundant information and low-cost. But the performance of visual sensors is highly affected by low light conditions when AVs driving at nighttime and in the tunnel. The low light conditions decrease the image quality and the performance of object detection, and may cause safety of the intended functionality (SOTIF) problems. Therefore, to analyze the performance limitations of visual sensors in low light conditions, a controlled light experiment on a proving ground is designed. The influences of low light conditions on the two-stage algorithm and the single-stage algorithm are compared and analyzed quantificationally by constructing an evaluation index set from three aspects of missing detection, classification, and positioning accuracy.

Gas diffusion layer (GDL), as a critical constituent of the proton exchange membrane fuel cell (PEMFC), plays a key role in mass, heat, electron, and species transport. GDL generally has two distinct layers: a macro-porous substrate (MPS) and a micro-porous layer (MPL). The fibers in MPS and the cracks formed during the deposition process on the surface of MPL change the overall transport capacity and effect the output performance of PEMFC. In this paper, methods of identifying the structural features of fibers and cracks in GDL images based on artificial intelligence are proposed. The block probabilistic Hough transform and the quadric voting based on the weighted K-means algorithm are programmed to realize the fiber feature extraction, and the crack feature extraction is realized by the regional connectivity algorithm and the geometric feature calculation based on the circumscribed graph of the crack region.

As the connectivity of vehicles increases rapidly, more vehicles have the capability to communicate with each other. Because Vehicular Ad-hoc NETworks (VANETs) have the characteristics of solid mobility and decentralization, traditional security strategies such as authentication, firewall, and access control are difficult to play an influential role. As a soft security method, trust management can ensure the security attributes of VANETs. However, the rapid growth of newly encountered nodes of the trust management system also increases the requirements for trust establishing mechanisms. Without a proper trust establishment mechanism, the trust value of the newly encountered nodes will deviate significantly from its actual performance, and the trust management system will suffer from newcomer attacks.

Lane marking detection plays a crucial role in Autonomous Driving Systems or Advanced Driving Assistance System. Vision based lane marking detection technology has been well discussed and put into practical application. LiDAR is more stable for challenging environment compared to cameras, and with the development of LiDAR technology, price and lifetime are no longer an issue. We propose a lane marking detection algorithm based on solid-state LiDARs. First a series of data pre-processing operations were done for the solid-state LiDARs with small field of view, and the needed ground points are extracted by the RANSAC method. Then, based on the OTSU method, we propose an approach for extracting lane marking points using intensity information.

A Joint Probabilistic Data Association (JPDA) multi-objective tracking improvement algorithm based on camera-radar fusion is proposed to address the problems of poor single-sensor tracking performance, unknown target detection probability, and missing valid targets in complex traffic scenarios. First, according to the correlation rule between the target track and the measurement, the correlation probability between the target and the measurement is obtained; then the measurement collection is divided into camera-radar measurement matched target, camera-only measurement matched target, radar-only measurement matched target, and no-match target; and the correlation probability is corrected with different confidence levels to avoid the use of unknown detection probability.

The main challenge for future autonomous vehicles is to identify their location and body pose in real time during driving, that is, “where am I? and how will I go?”. We address the problems of pose estimation and scene perception from continuous visual frames in intelligent vehicle. Recent advanced technology in the domain of deep learning proposes to train some learning models for vehicle’s series detection tasks in a supervised or unsupervised manner, which has numerous advances over traditional approaches, mainly reflected in the absence of manual calibration and synchronization of the camera and IMU. In the paper, we propose a novel approach for pose estimation and scene recognition with a deep fusion of multi-modal neural features in the manner of unsupervised. Firstly, low-cost camera and IMU are used to extract original visual and inertial data, then the visual and inertial encoders are utilized to encoder the feature of the two modes.

Since situational awareness is crucial for autonomous driving in urban environments, multi-target tracking has become an increasingly popular research topic during the last several years. For autonomous driving in urban environments, cars and pedestrians are the two main types of obstacles, and their motion characteristics are not the same. While in the current related multi-target tracking research, the same motion model (such as Constant Velocity model [CV]) or motion model set (such as CV combined with Constant Acceleration model [CA]) is mostly used to track different types of obstacles simultaneously. Besides, in current research, regular motion models are mostly adopted to track pedestrians, such as CV, CA, and so on, the uncertainty in pedestrian motion is not well considered.

Environmental perception is a crucial subsystem in autonomous vehicles. In order to build safe and efficient traffic transportation, several researches have been proposed to build accurate, robust and real-time perception systems. Camera and LiDAR are widely equipped on autonomous self-driving cars and developed with many algorithms in recent years. The fusion system of camera and LiDAR provides state-of the-art methods for environmental perception due to the defects of single vehicular sensor. Extrinsic parameter calibration is able to align the coordinate systems of sensors and has been drawing enormous attention. However, differ from spatial alignment of two sensors’ data, joint calibration of multi-sensors (more than two sensors) should balance the degree of alignment between each two sensors.

Lithium-ion batteries (LIBs) with relatively high energy, power density and eco-friendly characteristic are considered as a vital energy source in consumer market of portable electronics and transportation sector especially in electric vehicles (EVs). To meet the higher capacity requirements, the nickel-rich LIBs with higher capacity has been used as the commercial power batteries. However, the battery with higher energy density is more destructive, which could result in thermal runaway accidents and make the battery safety issues become more and more prominent. Thermal runaway of LIBs is one of the key scientific problems in safety issues. Until now, the inducement of thermal runaway process is complicated which perplex researchers and industry a lot. On the one hand, the internal mechanism about thermal runaway should be deeply studied. On the other hand, in-situ monitoring should be developed to supply the mechanism study and early warning.

In order to explore approaches for improving object detection accuracy in intelligent vehicle system, we exploit super-resolution techniques. A novel method is proposed to confirm the conjecture whether some popular super-resolution networks used for environmental perception of intelligent vehicles and robots can indeed improve the detection accuracy. COCO dataset which contains images from complex ordinary environment is utilized for the verification experiment, due to it can adequately verify the generalization of each algorithm and the consistency of experimental results. Using two representative object detection networks to produce the detection results, namely Faster R-CNN and YOLOv3, we devise to reduce the impact of resizing operation. The two networks allow us to compare the performance of object detection between using original and super-resolved images. We quantify the effect of each super-resolution techniques as well.

In the Kmeans cluster segmentation used in traffic scenes, there are often zone optimization and over-segmentation problems caused by the algorithm randomly assigning the initial cluster center. In order to improve the target extraction effect in traffic road scenes, this article proposes an improved Kmeans (IM-Kmeans) method. Firstly, search for the histogram peaks of the whole pixels based on hue, saturation, value (HSV) image, and find the initial cluster centers’ positions and number. Secondly, the noise points which are far away from the center pixel are removed, and then the pixels are classified into the nearest cluster center according to its value. Finally, after the clustering model reaches convergence, the area-clustering method is used for another classification to solve the over-segmentation problem. The simulation and experimental comparisons show that the IM-Kmeans algorithm has higher clustering accuracy than the traditional Kmeans algorithm.

The camera-based advanced driver assistance systems (ADAS) like lane departure warning system (LDWS) and lane keeping assist (LKA) can make vehicles safer and driving easier. Lane detection is indispensable for these lane-based systems for achieving vehicle local localization and behavior prediction. Since the vision is vulnerable to the variable environment conditions such as bad weather, occlusions and illumination, the robustness is important. In this paper, a robust algorithm for detecting and tracking multiple lanes with arbitrary shape is proposed. We extend the previously lane detection and tracking process from the space domain to the temporal-spatial domain by using a more robust and general multi-lane model. First, new slice images containing temporal information are generated from image sequences. Instead of binarization process, we use a more general detector for extracting the lane marker candidates with prior knowledge to generate the binary slice image.

In order to avoid the influence of the change of the camera pitch angle and the variation of the height of the ground on the ranging accuracy, improve the real-time performance of the algorithm by substituting the current widely-used monocular vision ranging algorithm which builds the optical model based on the shadow of the vehicle floor and the lane line, as well as avoid the classification of vehicle detection, a vehicle distance measurement algorithm based on monocular vision and license plate width is established. Firstly, the target image acquisition and preprocessing are studied. Then the paper studies the license plate image location segmentation method based on accelerated template matching. On this basis, the algorithm for obtaining the ratio of license plate width to image width is studied, and the function of vehicle distance and license plate ratio width is established.

This paper reports an experimental investigation of the wake flow behind a 1/12 scale notchback MIRA model at Re = UL/ν = 6.9×105 (where U is free-stream velocity, L the length of the model and ν viscosity). Focus is placed on the flow asymmetry over the backlight and decklid. Forty pressure taps are used to map the surface pressure distribution on the backlight and decklid, while the wake topology is investigated by means of 2D Particle Image Velocimetry. The analysis of the instantaneous pressure signals over the notch configuration clearly shows that the pressure presents a bi-stable behavior in the spanwise direction, characterized by the switches between two preferred values, which is not found in the vertical direction.