Huge amount of Traffic video segmented into manageable shots is the key step of database storage and video analysis in Intelligent Transportation Systems (ITS). Then key frames are extracted for representing main visual content of each shot. This paper proposes a novel approach for the segmentation of traffic video by the judgment of motion trend and supported by the vehicle status changes. Considering the number of sub-shots in shots as the initialized clusters number, we apply an improved global k-means clustering algorithm to extract the key frame. With the numerical experiments on traffic surveillance video using the propose method in this paper, shot boundary detection can be made in an effective manner. The extracted key frames by our approach also show better representation for the visual content of the video shot compared with other methods.
展开▼
