航母舰面保障作业全场景运动目标识别与跟踪技术研究

Carrier-based aircraft deck operation scheduling is one of the most important factors that influence the performance of the aircraft carrier. Carrier-based aircraft deck operation managing and controlling is one of the most important factors that influence the performance of the aircraft carrier. Conventional deck operation managing and controlling mainly adopts the manner of using interphone for communication and the crew on the deck reporting the state information of the whole deck operation to the command and control officers, which has low efficiency, and there is information delay, so that it cannot meet the demand for the efficient, precise, and safe deck operation under sustained high-intensity confrontation. Automation and intelligence of carrier-based aircraft deck operation managing and controlling is the trend of the aircraft carrier development. This research proposes to utilize the scene understanding technique based on machine vision to explore the key methods of detecting, tracking and visualizing the moving objects in the deck, including the aircraft, vehicle and crew. The research content mainly includes: (1) a novel method that utilizes the atmospheric scattering model to enhance and denoise the images adaptively; (2) novel methods that utilize generative adversarial and transfer learning to detect the objects on the deck, and novel methods that adopt multi-scale spatio-temporal context and multiple features fusion to track the objects across the cameras; (3) a novel method that uses the local feature and the optimum stitching plane. Through the above research, we aim at realizing the real-time detection, tracking and visualizing for the moving objects in the scene of carrier-based aircraft deck operation, and verifying the key technologies on the simulation system, in order to provide strong support to improve the intelligence level of the carrier-based aircraft deck operation scheduling of our country.

航母舰面保障作业管控是影响航母作战性能的关键因素之一。传统管控过程主要通过对讲机通信和保障人员汇报使指挥官掌握舰面保障作业全局状态信息的方式效率低下，存在信息延迟，难以满足持续、高强度对抗下高效、精确和安全的舰载机保障作业需求。舰载机保障作业管控的自动化和智能化是未来航母发展的必然趋势。本课题拟采用基于机器视觉的场景理解技术，探索实现舰面保障作业场景的运动目标舰载机、保障车辆和人员的检测、跟踪及其全景可视化的关键技术。研究内容主要包括：（1）基于大气散射模型的自适应图像增强和去噪；（2）基于生成对抗和迁移学习的舰面目标检测，以及基于多尺度时空上下文和多特征融合的跨摄像机目标跟踪；（3）基于局部特征和最佳拼接平面的跟踪视频图像全景拼接。通过上述研究，实现舰面保障作业场景运动目标的实时检测、跟踪及可视化，并将相关核心技术在仿真系统上验证，为提高我国航母舰面保障作业管控的智能化水平提供有力支持。

航母舰面保障作业管控是影响航母作战性能的关键因素之一。传统管控过程主要通过对讲机通信和保障人员汇报使指挥官掌握舰面保障作业全局状态信息的方式效率低下，存在信息延迟，难以满足持续、高强度对抗下高效、精确和安全的舰载机保障作业需求。舰载机保障作业管控的自动化和智能化是未来航母发展的必然趋势。本项目采用基于机器视觉的场景理解技术，探索实现舰面保障作业场景中的运动目标舰载机、保障车辆和人员的检测、跟踪及其全景可视化的关键技术，主要研究内容包括：（1）基于无监督学习的图像增强和去噪；（2）基于生成和多视图学习的舰面目标检测，以及基于多尺度时空上下文和多特征融合的跨摄像机目标跟踪；（3）基于局部和全局特征的全景跟踪视频图像拼接。通过上述研究，实现舰面保障作业场景运动目标的实时检测、跟踪及可视化，并将相关核心技术在仿真系统上验证，为提高我国航母舰面保障作业管控的智能化水平提供有力支持。本项目已经按照原定计划如期完成，项目执行期间一共发表较高质量学术论文6篇，其中包括IEEE-TCSVT、IEEE-TCDS、Neural Computing and Applications、Neural Processing Letters和 ACCV，并申请相关发明专利1项，目前处于公开审查状态，搜集并标注舰面保障作业场景运动目标数据集一个，有利于本方向继续开展研究。本项目的研究成果为基于视觉的舰面保障作业场景语义信息的智能化传递提供了有力的理论与方法支持，有利于提高舰载机保障作业管控的效率、准确性和安全性，能更好地支持航母舰面保障作业智能化管控。