Video Motion Event Detection Method Based On Scene And Motion Feature Classification

With the rapid development of network applications, the number of multimedia files increased sharply. The video itself is a kind of multimedia files, including a variety of media data like image, text, voice and etc. How to quickly deal with rich semantic information contained in video data, analysis and understand the video in a better way, have become a hot topic in nowadays.Video motion event describes the event of a kind of scene and motion object, so video scene selection can be firstly done for the video motion event detection of different scene and different classes object. On the basis of this, video event detection can reduce the search range, and improve the retrieval speed. For a number of different events, it can detect event through extracting video feature and then classifying event for annotation by machine learning. Therefore, the paper proposed a method based on scene and motion feature classification for event motion detection in video.The scene and motion feature classification method for video motion event detection contains two aspects: scene selection and motion event detection. According to different scenarios, it offered a scene selection algorithm based on video static feature. The video static feature got through selecting a number of key frames from the video, extracting color feature and HOG feature from each of the key frames, and fusing the two features in a linear way, and then using the linear weighted method to fuse features of multiple frames. Then support vector machine was used for video scene selection. For a number of motion events, it proposed a motion event detection algorithm based on video motion feature. The first step was to get the tracks by track extraction. Secondly, it did a series of pretreatments as noise elimination, track number selection and etc. And then handled the trajectory through segmentation algorithm based on the threshold. According to the question of multi-sequence and different length of video trajectory, it presented a method of multiple sequence distance based on minimum distance. On the basis of it, the paper respectively used the K nearest neighbor and support vector machine algorithm to classify and annotate the video motion event for event detection. Finally, the paper proved the effectiveness of the scene and motion feature classification method for video motion event detection through experiments.