Mobile technology platforms today possess various integrated sensors, such as smartphones and tablets, which could be used as an alternative platform for estimating vehicle dynamics and recognizing driving events. However, the main challenge for implementing effective vehicle/driver monitoring platforms in real-world environments is knowledge of the orientation and relative movement of the devices inside the vehicle. In this study, we first present the framework of MobileUTDrive-App, an Android App that developed for in-vehicle data collection which mimics the actual UTDrive instrumented vehicle from UT Dallas. The motivation of using a portable device for driving record collection will be described, followed by a discussion of system design in hardware and software, as well as details of software implementation. Next, an evaluation study is presented which further utilizes our previous proposed method of converting device-referenced inertial measurement unit (IMU) readings into vehicle-referenced accelerations, which allows for a free-positioned device for in-vehicle dynamics sensing. A driving event recognition task is also conducted to evaluate the performance of the proposed method in real-world environments. By taking advantage of deep learning approaches, a comparable result is observed between our proposed method and a fixed-positioned device baseline. After discussion, a final implementation processing pipeline is suggested for development of future real-time systems.
The communication inside a vehicle can be impaired by background noise while driving at medium or higher velocities and by the unusual orientation of the passengers as they are not facing each other. As a consequence, the passengers often start to raise their voices and to change their positions by leaning forward or, in case of the front passengers, turning backwards to improve this situation. These adaptations to the communication situation may be uncomfortable, exhausting, and might result in an increased risk for accidents if the driver turns around (and does not look on the street). An in-car communication system (ICC system) can improve the communication by recording the speech signals and reproducing the enhanced signals over the loudspeakers inside the compartment, which are close to the listening passengers. After and during the design and the development of such systems, the achieved quality is of great interest. Therefore, a spectral distance approach is presented in this chapter. Even if spectral distance measures are well established and often used for quality evaluation in speech communication systems, we face here a special challenge, which is the creation of an appropriate reference signal. In addition, an investigation of the reliability of the proposed quality estimate by means of a subjective listening test was performed and the results are presented.
The communication between passengers in a car cabin is often disturbed by background noise even when driving at only medium speed. One solution to improve speech intelligibility and thus usually also communication quality is the usage of so-called in-car communication (ICC) systems. Such systems record each passenger᾿s voice using microphones that are installed in the car. After several stages of low-delay signal processing, the enhanced signals are played back using the car᾿s loudspeakers. Even if the basic concept is rather simple, such a system faces several problems. The most important one is feedback, which occurs when the sound, that is played back, is recorded again with the same microphone. This closed electro-acoustic loop can lead to instability when the system gain is too large. To overcome this problem different solutions exist. One is to use adaptive feedback cancellers which estimate the acoustic paths between the loudspeakers and the microphones. This is very similar to acoustic echo cancellation (AEC) in hands-free systems that can be found in most modern cars. In addition, the hardware components might be the same which then results in the fact that both, the hands-free and the ICC system, are estimating the same acoustic paths. In this work, we propose a method to combine both kinds of systems to a signal enhancement system with both hands-free and ICC capability. We show which adaptations have to be applied to each of these systems to make them “cooperative,” numerically efficient, and robust against distortions. Additionally, methods for the control and the adaption of the adaptive filter will be shown.
Several automotive noise power spectral density (PSD) estimation algorithms have been developed over the years in the context of speech enhancement. However, one of the most practical and in practice often used methods - the multiplicative constant-based method - has received only little attention. In this chapter an attempt is made to formalize this method. At first, the noise estimation approach is presented as a set of general rules, which is applicable to any algorithmic realization of multiplicative constant-based methods. The generalized rules are then applied to create a noise PSD estimation algorithm. This algorithm is further analyzed by considering various (sub-)stages of the algorithm where it is empirically shown that the behavior conforms to the design and eventually relate to the generalized rules. Finally, a wide set of results show the performance in comparison with two popular algorithms developed over the years by two independent research groups.
In-car communication (ICC) systems enhance speech communication between passengers inside a car cabin. They take up the front row’s speech signal with a microphone and remove acoustic feedback components. The enhanced microphone signal is then additionally amplified and played back via the rear loudspeakers, thereby increasing speech intelligibility for the rear seat row. In an extended set of relevant acoustic scenarios for ICC systems, also including an additional FM radio signal, we investigate and discuss the frequency domain adaptive Kalman filter as mono-, stereo-, and quadraphonic-channel acoustic feedback cancellation under the very same acoustic conditions.
The rapid advances in sensing and computing technologies in recent years have made the use of self-driving cars in our everyday lives seem like a more and more achievable goal. As open-source software platforms for autonomous driving have matured, and the required sensors fall in price, conducting research in self-driving has become more easily accessible. Aimed at further reducing the barrier to entry of developing self-driving technologies, the PIX Moving “KuaiKai” was held as a self-driving hackathon in Guiyang City, Guizhou, China, between May 20 to 27, 2018. Armed with two electric vehicles and a variety of off-the-shelf sensors, thirteen entrants combined their efforts to build a self-driving car to race with a human driver on a pre-defined circuit, with a set of everyday driving tasks and obstacles to overcome. This chapter describes the competition structure and the technical details of the systems which were designed and implemented. Although it was constructed in only seven days, the final vehicle was capable of completing the circuit autonomously and could perform a number of localization, perception, and path planning tasks.
Nowadays, the development of driving support systems and autonomous driving systems have become active. Vehicle ego-localization using in-vehicle sensors is one of the most important technologies for these systems. Accordingly, various attempts to localize own vehicle from in-vehicle sensors have been made. In general, the estimation accuracy of the traveling direction is lower than in the lateral direction. Therefore, we present a highly accurate method for ego-localization of the traveling direction based on epipolar geometry using an in-vehicle monocular camera. The presented method makes correspondences between in-vehicle camera images and database images with location information, and calculates the location using locations annotated to the corresponding database images. However, there are many gaps due to the difference in speed and trajectory of vehicles even if the images are obtained along the same road. To overcome this problem, the distance between the input image and the database image is calculated by the distance metric based on the epipolar geometry and the local feature method. An experiment was conducted using actual images with correct locations, and the effectiveness of the presented method was confirmed from its results.
Connected and automated vehicles combine leading edge technologies to allow vehicles to be self-aware, and communicate roadway information with other vehicles and drivers. Among the most promising applications for connected and automated vehicles (CAVs) is platooning, that is, the synchronized movement of two or more vehicles as a unit. Platooning holds great potential to make road transport safer, cleaner, and more efficient. In this chapter, we discuss CAV technologies, architectures, and computing approaches for platooning.
The sixteen chapters in this Volume 2 (IVT 2) of the book series entitled Intelligent Vehicles and Transportation Systems, from the contributing authors have focused on the technologies of data analytics in speech/audio, computer vision, and intelligent vehicle advancements under the perspectives of human harmonizing systems such as human- machine interactions (speech understanding, synthesis, virtual reality, gesture, dialogues, multimodality, etc.), computer vision (image recognition, tracking, segmentation, depth maps, visual simultaneous localization and mapping (SLAM)), and intelligent vehicles (driver behaviour, path planning, sensor fusion, point clouds, dynamic maps). The contributions were grouped into three parts. The first group consisting of eight chapters (Part A) have addressed the driver/vehicle interaction systems. The second group of five chapters (Part B) were on models and theories of driver/vehicle systems. The final four chapters made up Part C and discussed self-driving system technologies.