基于深度学习与视觉感知的机场场面监视技术研究

Airport scene surveillance is an important technical method to ensure the safe and efficient operation of airports. Traditional scene surveillance methods are either too costly or difficult to surveil non-cooperative targets. The application of vision perception in airport scene surveillance is a low-cost and non-cooperative surveillance solution that uses video surveillance data acquired by cameras to surveil and analyze various types of airport scenes. However, the existing video analysis methods have higher requirements on image sharpness, and have great limitations in various weather conditions, illumination conditions and target size etc, making it difficult for the video surveillance system to ensure full-time, reliable and autonomous operation. In order to solve the above problems, this project focuses on the application of essential technologies in airport scene intelligent surveillance such as deep learning, generative adversarial network, multi-modal image fusion, and 3D spatial-temporal convolution etc. What’s more, our research focuses on studying methods of image enhancement and super-resolution video reconstruction in complex environments, which improve the quality of video surveillance data. Our research also focuses on studying the object detection and tracking methods based on efficient feature extraction of small regions to solve the tough problems of airport, including large-scale airport scene, small target and difficult target orientation. On this basis, our research focuses on studying target behavior recognition methods based on 3D spatial-temporal separation that analyze the behavior of targets in airport scenes, and provide early prediction and post-inspection for the illegal behaviors of targets through analyzing the behavior of targets in airport scenes.

机场场面监视是保证机场安全高效运行的重要手段，传统场面监视手段要么成本过高，要么难以监视非协作式目标。将视觉感知技术应用于机场场面监视是一种低成本的非协作式监视方案，即利用摄像机获取到的视频数据，对机场场面的各类目标进行监控与分析。但现有视频分析方法对图像清晰度的要求较高，在天气条件、光照条件、目标大小等方面的适应性存在较大的局限，使视频监视系统难以保证全天候、可靠地自主运行。为解决以上问题，本课题致力于深度学习、生成对抗神经网络、多模态图像融合、三维时空卷积等关键技术在机场场面智能监视中的应用，研究复杂环境下图像增强及视频超分辨重构方法，提升视频数据的质量；研究基于高效小目标区域特征提取的目标检测与跟踪方法，以解决机场场面大、目标小、目标定位困难的问题；在此基础上，研究基于三维时空分离的目标行为识别方法，以分析机场场面中目标的行为，为机场场面中目标违规行为的提前预知和事后回查提供依据。

机场场面监视是保证机场安全高效运行的重要手段，传统场面监视手段要么成本过高，要么难以监视非协作式目标。将视觉感知技术应用于机场场面监视是一种低成本的非协作式监视方案，即利用摄像机获取到的视频数据，对机场场面的各类目标进行监控与分析。但现有视频分析方法对图像清晰度的要求较高，在天气条件、光照条件、目标大小等方面的适应性存在较大的局限，使视频监视系统难以保证全天候、可靠地自主运行。为解决以上问题，本课题致力于深度学习、生成对抗神经网络、多模态图像融合、三维时空卷积等关键技术在机场场面智能监视中的应用，研究复杂环境下图像增强及视频超分辨重构方法，提升视频数据的质量;研究基于高效小目标区域特征提取的目标检测与跟踪方法，以解决机场场面大、目标小、目标定位困难的问题;在此基础上，研究基于三维时空分离的目标行为识别方法，以分析机场场面中目标的行为，为机场场面中目标违规行为的提前预知和事后回查提供依据。

内容获取失败，请点击重试