铁路入侵运动目标实时检测技术

针对铁路场景下入侵异物的特点，采用智能视频技术，对监控视频图像序列中入侵运动目标检测方法进行研究．提出基于参考点的“相对背景差分法”、基于目标特征的跟踪算法和基于透视规律的目标分类方法，实现对多目标场景运动目标的实时检测识别。典型场景实验结果表明：上述算法实现了铁路入侵运动目标的高效检测，与基础背景差分法相比，误检率和漏检率分别减小了24．56％和54．17％；与基于区域的传统目标跟踪方法相比，误匹配率和漏匹配率分别减小了64．78％和22．58％，且算法具有较强的实时性和鲁棒性。

Moving target real- time detection of railway intrusion

According to the characteristics of the intruding objects in railway scene, detection method for the moving target in the video image sequence was studied based on intelligent video technology. To realize real - time detecting and recognizing of moving targets in multi - object scene, a relative background difference method based on reference points, a tracking algorithm based on features and a classification method based on the laws of perspective were proposed. The results of typical scenes show that the above - mentioned methods can attain high efficiency detection of railway intrusion. The average missed detection rate and the false detection rate are de- creased by 24.56% and 54.17% respectively compared with that of the traditional background difference meth- od. The average error - matching rate and miss - matching rate are decreased by 64.78% and 22.58% respec- tively in comparison with that of the traditional region matching algorithm. The proposed method is significantly efficient and robust.