Traffic surveillance video data extraction and indexing

This research provides a practical method for extracting traffic parameters from raw traffic surveillance videos. We will use the extracted traffic information and provide a video indexing method that can be used for content-based queries from raw traffic videos. In our data extraction process, we will introduce an object-based video segmentation algorithm. Most of the video segmentation methods are currently based on shots. Our segmentation relies on objects in the video instead of frame differences. We will introduce a novel dynamic region-based tracking algorithm and develop an illumination invariant comparison technique for noise reduction in object detection. This generic process reduces sensitivity of the object detection to lighting conditions, which cannot be controlled in any surveillance videos. The extracted data are calibrated to actual distances through a background image calibration process. The calibration method involves a manual procedure and relies on data from roadway segment in the video. Traffic parameters are, then, calculated for each object at each frame. Objects data are converted to frame information. We, finally, create indexes between object-based traffic parameters and video frames. We will show how this method of video indexing can be used in content-based queries from raw traffic videos and how to reduce storage space for traffic surveillance videos. The experiments show that our approach can provide accurate vehicle detection and traffic parameters extraction, high correlation between observed and actual data, and an efficient archiving alternative for storing the raw video that has a smaller file size without losing any information of interest.