Automatic Guided Veh的問題，透過圖書和論文來找解法和答案更準確安心。 我們找到下列包括價格和評價等資訊懶人包

緊急鳴笛車輛之自動偵測方法研究

為了解決Automatic Guided Veh的問題，作者TRAN VAN THUAN 這樣論述：

Emergency vehicles (EV), such as fire trucks, police cars, and ambulances, are the crucial components of the emergency service system (EMS), which provides quick responses and professional aids in urgent situations. For example, in case of reported serious illness or injury, the focus of EMS, inclu

ding ambulance, is providing rapid transportation and emergency medical care of the patients. Due to the need of driving at high speed to reach the destination, the EV’s drivers may put themselves at risk. Furthermore, in certain driving scenarios, car drivers may sometimes be unaware of the approac

hing EV, for instance, when an EV’s siren is unclear due to the use of the in-vehicle audio system, or when an EV is out of the car driver’s vision, so non-emergency vehicles may block or even collide with the EV. This work studies automatic methods for emergency vehicle detection (EVD) to warn car

drivers of the nearby priority vehicle(s) and paying attention.This dissertation investigates audio-based and vision-based approaches to build deep learning-based EVD systems that can accurately detect the EV using their siren sounds and/or their visual presence. Firstly, we build different convolut

ional neural networks (CNNs) for two kinds of EVD systems, namely A-EVD and V-EVD, which are based on siren sound detection and object detection approaches, respectively. Then, we integrate models from A-EVD and V-EVD to develop a prototype of the audio-vision EVD system (AV-EVD). To our knowledge,

there is no prior work examining such an AV-EVD system. In A-EVD, besides investigating the combined use of acoustic handcrafted features, including MFCCs and log-mel spectrogram, to train the 2D-CNN model (MLNet), we propose to train the end-to-end network (WaveNet) directly on audio raw waveforms.

Our experiments on a custom dataset of three audio classes (i.e. siren sound, vehicle horn, and traffic noise) show the efficiency of the proposed handcrafted feature aggregation as well as the raw feature extraction methods. Also, we propose two-stream models, namely PreCom-SirenNet, PostCom-Siren

Net, and DF-SirenNet, which are trained on both handcrafted features and raw wave features to further boost the classification accuracy. Our proposed A-EVD models can work well with various input lengths between 0.25 seconds and 1.5 seconds to obtain accuracies ranging from 92.2% to 98.51%. In V-EVD

, we propose to apply and modify the YOLOv4 object detection algorithm to build a single-stage V-EVD system, namely YOLO-EVD, which achieves 95.5% mean average precision on our custom dataset. The AV-EVD system comprised of YOLO-EVD and the WaveResNet, an improved version of the WaveNet, also yields

promising results, showing the potential of fusing information from acoustic signals and visual information to enhance the reliability of the system’s predictions. The application of A-EVD, V-EVD, and AV-EVD from this work will not only be able to help drivers avoid car accidents, but also provide

a necessary safety function for other smart vehicles and traffic infrastructures, such as self-driving cars and intelligent traffic light control systems.

非完整限制之移動機器人的恢復計劃

為了解決Automatic Guided Veh的問題，作者Kothandaraman Kannadasan 這樣論述：

The thesis presents a recovery planning in different environments for a non-holonomic robot. The classical navigation system allows robots to navigate from one place to another in a collision-free manner. Unfortunately, a stuck condition occurs when facing the closest obstacle. When the robot encou

ntered the closest obstacle, the robot's footprint touched the obstacle area in the costmap, and at that point, it gave up and failed to overcome. In this work, we propose a backward recovery behavior for a non-holonomic robot which improves a mobile robot’s navigation and improves the pose of the r

obot always towards the goal point. The experimental results are presented in order to verify the feasibility and usefulness of the proposed behavior algorithm. The thesis finishes with a comprehensive system demonstration in simulation utilizing the Turtlebot3 mobile robot running under ROS Kinetic

.