Video Motion Detection Based On Optical Flow Field

Image has a clear, simple and direct trait that makes it easy understood. It goes: "listening a lot is not more than a glance", "one picture overpasses many words". Video images are composed by a series of frames and frame owns most character which is originally rich data, strong correlations between nearly images and dynamically changed mode in time field, which make detection, segmentation and recognition possible. However, the object of detection, segmentation and recognition has much important application not only to industry and military affairs, but also to docks, public, banks, and vehicle surveillance system. Therefore research on video image has a great significance to use.This paper aims on motion detection based on static background. Most important content studies character-based video image filter which is maximum and minimum according to median. Investigate and realize canny operator that has high and low threshold. Depending on eight directions neighborhood chain code, and using clockwise and anticlockwise labeling, raise a new way to form filling.Paper also analyses optical flow limited equation and researches motion estimation operators including Horn-Schunk optical flow, patch matching optical flow and iterative optical flow. Realizes and compares above optical flow operators.At the same time, realize and contrast conventional snake operators and gradient vector flow snake in detail and improve gradient vector flow operator to establish multi-scale gradient vector flow operator.Based on Horn-Schunk optical flow and multi-scale gradient vector flow snake, the paper sets up a moving detection system on video image character: mainly includes pre-processing, time-space fuse, motion estimation, snake growing and background renew. At last, applying these processes to video images including hall monitor under complicated illumination conditions, it has the practical meaning in application and experience proves all the operators in thepaper.