An Efficient Key Frame Extraction from Surveillance Videos for Real-World Anomaly Detection

Technology development has entrenched video analysis in numerous applications. Video analysis is important to obtain the data contents rather than its characteristics. Instead of examining the entire data content analyzing the essential portions as key frame reduces the computation time and memory. Extracting the important features from the dataset is termed as key frame extraction and this research work presents an efficient methodology to extract important frames from the video instances using HSV histogram and k-means clustering algorithm. The raw video data is preprocessed and then data points are clustered using k-means clustering. The proposed k-means clustering based key frame extraction technique is experimented with using UCF Crime dataset and the key frames are extracted from seven categories of videos. Detailed analysis of actual features and extracted features are comparatively analyzed and the observations are presented in the experimental section.