面向动摄像机的高速运动目标检测

Detecting high-speed moving targets in moving camera environments

目的为解决动摄像机中高速运动目标检测复杂度高的问题,提出一种基于压缩视频运动矢量的高速运动目标检测新方法。方法该方法首先分析监控视频的码流格式和解码特点;然后从视频流中直接提取运动矢量;接着进行运动矢量规范化,并根据3σ准则提取场景的全局运动参数;最后通过对运动矢量统计特征的分析,实现面向动摄像机的高速运动目标快速检测。结果仿真实验表明,该方法在经典和自建数据库上目标提取效率较现有算法均有较大提高。结论本文方法充分利用了压缩视频数据中蕴含的运动信息,极大降低运动目标检测的复杂度,可以有效提取动摄像机成像画面中的高速运动目标,在经典和自建数据库上的目标提取效率较现有算法均有较大提高。

Objective Detecting moving targets in moving camera environments from video stream is a fundamental step in many computer vision applications, such as intelligent visual surveillance, human-machine interaction, and content-based video coding. Background subtraction is generally regarded as an effective method for extracting foreground objects. Given the existence of extensive literature regarding background subtraction, most existing methods assume a stationary camera. This assumption limits the applicability of these methods to moving camera scenarios. Background is a complex environment that usually includes distracting motions, which make the task more challenging. In addition, with the development of coding techniques, high-definition videos have become widely used. Thus, extracting and updating background images become complex and time-consuming processes. To solve this problem, a fast target detection method is proposed on the basis of motion vectors. Method The data format and decoding features of surveillance videos are analyzed in this study. Subsequently, methods by which to obtainmotion vectors directly from a video stream are determined, and results confirm the method validity by comparing the motion vectors extracted through this method with those extracted using the H. 264 Visa business software. Moreover, the motion vectors are normalized by considering the reference frames, thus preventing the emergence of singular motion vectors and making the distribution of the motion vectors more reasonable. Global motion is detected according to 3σ theory, and different compensation algorithms are proposed for various scenarios. Finally, targets are extracted by analyzing the statistical property of motion vectors. Result Detecting moving targets from a moving camera is difficult. For conventional methods, moving background is not always considered, and systems that have a delayed response to background changes are unsuitable for this situation. To meet the real-time performance requirements of the system, motion vectors are used to detect global motion and extract motion targets, thus reducing redundant computation effec- tively. Test results show that the proposed method is superior to conventional methods. Conclusion To prove the validity of this algorithm, an extensive set of experiments are performed, and the proposed method is compared with some of the most recent approaches in the field by using publicly available datasets and a new annotated dataset. Results prove that the proposed approach can effectively extract high-speed moving targets from a moving camera.