基于视频深度学习的铁路周界入侵检测算法研究

基于视频智能分析的铁路周界入侵检测算法相比于雷达、振动光纤，具有成本低、误报率低的优点. 针对视频中存在不同分辨率目标的问题，提出一种改进的Cascade Mask RCNN( CMR)模型，使用级联结构获得目标的准确定位. 为增强模型对小目标的检测能力，在原始模型的基础上，增加基于特征金字塔网络(FPN)的多尺度特征提取模块和基于空洞金字塔汇聚(ASPP)子网络的空间上下文增强模块. 在实际铁路周界入侵场景视频中验证了模型的有效性. 结果表明，该模型可实现不同场景下的铁路周界入侵检测，相较于原始模型，新模型对小目标检测的F-measure 提高了0.24. 模型既解决了不同场景下铁路周界入侵检测问题，又有效地提高了视频智能分析对小目标检测的准确率.

Railway Perimeter Intrusion Detection Algorithms Based on Video Deep Learning

Compared with radar and vibration optical fiber, the railway perimeter intrusion detection algorithm based on video intelligent analysis has the advantages of low cost and low false alarm rate. Aiming at the simultaneous detection of large and small targets in videos, this paper proposes an improved Cascade Mask RCNN (CMR) model. The model uses the cascaded structure to locate the target accurately. At the same time, a multi- scale feature extraction model is added to the original model to enhance the expressive ability for small targets. Multi-scale module is implemented by feature pyramid networks (FPN). The spatial context enhancement module is implemented through an atrous spatial pyramid pooling (ASPP) subnetwork. The effectiveness of the proposed model was verified by the actual railway scene videos. The results show that the new model improves the F-measure of small targets detection by up to 0.24 compared with the original model. The proposed model enhances the detection capability of railway perimeter intrusion in different scenarios, and improves the accuracy of video detection for small targets.