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. 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 3 theory

and different compensation algorithms are proposed for various scenarios. Finally

targets are extracted by analyzing the statistical property of motion vectors. 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 effectively. Test results show that the proposed method is superior to conventional methods. 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.