CPS: Small: Robust and Efficient Perception System for Autonomous Vehicles (REPAVE)

CPS：小型：用于自动驾驶车辆的稳健且高效的感知系统 (REPAVE)

• 批准号: 1931867
• 负责人: Mooi-Choo Chuah
• 金额: $ 50万
• 依托单位: Lehigh University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-11-01 至 2023-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1931867&HistoricalAwards=false
• 关键词: CPS; Small; Robust; Efficient; Perception

The goal of this project is to advance the science of designing efficient and robust perception systems for complex cyber-physical systems such as autonomous vehicles (AV). Industries in the US, especially AV-related industries, must be competitive in terms of innovations in computer vision and security issues related to deep learning models deployed in complex cyber-physical systems. This project provides these innovations and prepares new generations of computer scientists that can join the industry workforce. The solutions created in this project will ensure safer operations of autonomous systems which benefit the society overall. Finally, hands-on training sessions will be organized at top computer vision conferences to provide opportunities to members of underrepresented groups. This project explores innovative solutions for efficient deep learning models in computer vision which are suitable for resource constrained devices and smarter sensors which can improve the robustness of the perception systems in complex cyber-physical systems. First, a more robust deep learning based perception system utilizing a combination of a color camera and cheaper sensors with lower cost but with better accuracy and efficiency will be created. Second, autonomous systems typically operate under dynamic environments. A more robust decision making module which utilizes context information involving interactions among nearby moving agents to make quick and accurate predictions of their future movements will also be designed. Complex cyber-physical systems are subjected to cyberattacks, and smarter sensors with unique signatures will be designed to allow such perception systems to be more resilient.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

该项目的目标是推进为自动驾驶汽车（AV）等复杂的网络物理系统设计高效和强大的感知系统的科学。美国的行业，特别是与AV相关的行业，必须在计算机视觉创新和与部署在复杂网络物理系统中的深度学习模型相关的安全问题方面具有竞争力。该项目提供了这些创新，并准备新一代的计算机科学家，可以加入行业的劳动力。该项目中创建的解决方案将确保自主系统的安全运行，从而使整个社会受益。最后，将在顶级计算机视觉会议上组织实践培训课程，为代表性不足的群体成员提供机会。该项目探索了计算机视觉中高效深度学习模型的创新解决方案，适用于资源受限的设备和更智能的传感器，可以提高复杂网络物理系统中感知系统的鲁棒性。首先，将创建一个更强大的基于深度学习的感知系统，该系统利用彩色摄像头和成本更低但精度和效率更高的廉价传感器的组合。其次，自主系统通常在动态环境下运行。 一个更强大的决策模块，利用上下文信息，涉及附近的移动代理之间的相互作用，使他们的未来运动的快速和准确的预测也将被设计。复杂的网络物理系统容易受到网络攻击，具有独特签名的智能传感器将被设计成使这种感知系统更具弹性。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

UCTNet: Uncertainty-Aware Cross-Modal Transformer Network for Indoor RGB-D Semantic Segmentation

• DOI: 10.1007/978-3-031-20056-4_2
• 发表时间: 2022
• 作者: Xiaowen Ying;M. Chuah

Robustness of Trajectory Prediction Models Under Map-Based Attacks

基于地图的攻击下轨迹预测模型的鲁棒性

• DOI: 10.1109/wacv56688.2023.00452
• 发表时间: 2023
• 期刊: IEEE
• 作者: Zheng, Zhihao;Ying, Xiaowen;Yao, Zhen;Chuah, Mooi Choo
• 通讯作者: Chuah, Mooi Choo

Weakly-supervised Object Representation Learning for Few-shot Semantic Segmentation

• DOI: 10.1109/wacv48630.2021.00154
• 发表时间: 2021-01
• 期刊: 2021 IEEE Winter Conference on Applications of Computer Vision (WACV)
• 作者: Xiaowen Ying;Xin Li;M. Chuah

SRNet: Spatial Relation Network for Efficient Single Stage Instance Segmentation in Videos

SRNet：用于视频中高效单阶段实例分割的空间关系网络

• 发表时间: 2021
• 期刊: ACM Multimedia 2021
• 作者: Ying, Xiaowen;Li, Xin;Chuah, Mooi Choo
• 通讯作者: Chuah, Mooi Choo

Delving into Light-Dark Semantic Segmentation for Indoor Scenes Understanding

• DOI: 10.1145/3552482.3556556
• 发表时间: 2022-10
• 期刊: Proceedings of the 1st Workshop on Photorealistic Image and Environment Synthesis for Multimedia Experiments
• 作者: Xiaowen Ying;Bo Lang;Zhihao Zheng;M. Chuah