Abstract:  Event detection is defined as interesting events which attract the attention of users. Now a lot of research about video event detection depends strongly on specific domain knowledge and prior model result in it is
difficult to apply to other domain or even other database.The aim of this paper is to address recognition of natural events in diverse and realistic video settings. We propose a novel method based on bag of local spac}time interest
points’features to recognize and retrieval complex events in real movies. In this method，an individual video sequence is represented as a bag of local spacrtime features then we integrate such bag-of-feature with support
vector machine for recognition events. Local spacitime features have become a popular video representation for event recognition. Local spacrtime features capture characteristic shape and motion in video and provide relatively
independent representation of events with respect to their spatio-temporal shifts and scales as well as background clutter and multiple motions in the scene.The calculation of local representations proceeds in a bottorrrup fashion;
spatio-temporal interest points arc detected first，and local patches arc calculated around these points. Finally, the patches arc combined into a final representation. Spacrtime interest points arc locations in space and time where
changes of movement occur in the video. It is assumed that these locations arc more informative for recognition. So local space-time features arc introduced to capture the local events in video and they can also be adapted to size and
velocity of the pattern of the event. Since the number of frames of each video clip is different and the quantity of the spacrtime interest points of each frame is also diverse, so in order to unify the representation of video sequence, we
adopt the idea of bag of features. For every event class local spatio-temporal features used for training arc first quantized into visual words and a video is then represented as the frequency histogram over the visual words.To
evaluate effectiveness of this method，this paper uses the Hollywood dataset，in this dataset the shot sequences has collected from 32 different Hollywood movies, where the training samples come from 12 movies and the testing
samples arc from another 20 movies and it includes 8 event classes; answering the phone，getting out of the car, hand shaking，hugging, kissing, sitting down, sitting up and standing up.The presented result justify the proposed
method explicitly improve the average accuracy and average precision compared to other relative approaches.