Scholarlink

<p>Intelligent Transportation Systems (ITS) aim to enhance road safety and Internet of Things (IoT)-related solutions are crucial in achieving this objective. By leveraging Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) technologies, drivers can access valuable information about their surroundings. This research utilized the Unity 3D game engine to simulate various traffic scenarios, exploring a stochastic environment with two data sources: camera and road sign labels. We developed a full-duplex communication system to enable the communication between Python and Unity. This allows the vehicle to capture images in Unity and classify them using Convolutional Neural Network (CNN) models coded in Python. To improve road sign detection accuracy, we applied multi-sensor Data Fusion (DF) techniques to fuse the information received from the sources. We applied DF methods such as the Kalman filter, Dempster-Shafer theory, and Fuzzy Integral Operators to combine the two sources of information. Furthermore, our proposed CNN model incorporates an Ordered Weighted Averaging (OWA) layer to fuse information from three pre-trained CNN models. Our results show that the proposed model integrating the OWA layer achieved an accuracy of 98.81%, outperforming six state-of-the-art models. We compared the Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF). In our work, EKF exhibited a lower execution time (0.02 seconds), yielding less accurate results. UKF, however, provided a more accurate estimate while being more computationally complex. Furthermore, the Dempster-Shafer model showed approximately 30% better accuracy compared to the Fuzzy Integral Operator. Using this methodology on autonomous vehicles in our virtual environment led to making more accurate decisions, even in a variety of weather conditions and accident scenarios. The findings of this research contribute to the development of more efficient and safer vehicles.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Applications of pre-trained CNN models and data fusion techniques in Unity3D for connected vehicles

  • 期刊论文

  • 发表日期：

    2025年1月30日

查看全文
  • Mojtaba Norouzi,
  • Seyed Hossein Hosseini,
  • Mohammad Khoshnevisan,
  • Behzad Moshiri

摘要

Intelligent Transportation Systems (ITS) aim to enhance road safety and Internet of Things (IoT)-related solutions are crucial in achieving this objective. By leveraging Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) technologies, drivers can access valuable information about their surroundings. This research utilized the Unity 3D game engine to simulate various traffic scenarios, exploring a stochastic environment with two data sources: camera and road sign labels. We developed a full-duplex communication system to enable the communication between Python and Unity. This allows the vehicle to capture images in Unity and classify them using Convolutional Neural Network (CNN) models coded in Python. To improve road sign detection accuracy, we applied multi-sensor Data Fusion (DF) techniques to fuse the information received from the sources. We applied DF methods such as the Kalman filter, Dempster-Shafer theory, and Fuzzy Integral Operators to combine the two sources of information. Furthermore, our proposed CNN model incorporates an Ordered Weighted Averaging (OWA) layer to fuse information from three pre-trained CNN models. Our results show that the proposed model integrating the OWA layer achieved an accuracy of 98.81%, outperforming six state-of-the-art models. We compared the Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF). In our work, EKF exhibited a lower execution time (0.02 seconds), yielding less accurate results. UKF, however, provided a more accurate estimate while being more computationally complex. Furthermore, the Dempster-Shafer model showed approximately 30% better accuracy compared to the Fuzzy Integral Operator. Using this methodology on autonomous vehicles in our virtual environment led to making more accurate decisions, even in a variety of weather conditions and accident scenarios. The findings of this research contribute to the development of more efficient and safer vehicles.