Event Detection from Video Surveillance Data Based on Optical Flow Histogram and High-level Feature Extraction

This paper presents a new approach for event detection from video surveillance data based on optical fow histogram with no prior knowledge of the motion nature. First,we start by estimating the motion from images sequence using optical flow technique. Second, we perform a classification using the histogram of the optical flow vectors and we use a chain coding algorithm that we applied to each class for the spatial segmentation. Finally, we extract a high-level feature from any frame for use in the learning and search events by SVM and HMM. We have tested the developed method on real image sequences, our results are very promising.

I. Introduction

The proliferation of cameras in video surveillance systems creates a flood of information increasingly important, which is difficult to analyze with standard tools. If the image processing techniques currently used to detect abnormal events (such as cars against the sense objects abandoned), the search video data in a post, for example to find events of interest, represents a huge task if it must be done manually. Therefore, it is necessary to develop solutions in order to aid the search in a sequence of video surveillance. In this type of sequence, the user does not scenes or visuals, as in the case of excavation in movies or video archive, but rather events. The main use cases of search in data from video surveillance are: •

Detection and search of special events.

•

To lead and/or optimize the methods of classification and automatic detection of abnormal events.

•

For the management of infrastructure, for example; roads, access to shopping malls and public spaces.