Developing an image sensor system (ISENS) for real-time traffic surveillance using online CCTV videos.

摘要

Closed Circuit Television (CCTV) cameras are used on major highways and intersections for traffic surveillance and incident monitoring. These jobs are done manually inmost cases. The difficulty in automatically obtaining traffic information from CCTV videos is that the existing video/image processing methodologies cannot be applied directly to CCTV videos because of the camera's unknown calibration parameters and its movable features such as zoom, tilt, and pan.; This research developed an image sensor system (ISENS) which makes qualitative decisions about the current roadway traffic situation, such as, light traffic, heavy traffic, or congested traffic. The system was developed with videos from a, CCTV camera on the Internet. ISENS consists of four modules: background generation; bidirectional roadway detection; vehicle tracking, and traffic data acquisition. One quality and two quantity measures were defined to evaluate the performance of ISENS and assess roadway traffic conditions. Estimates of space occupancy, time headway, and flow rate of each direction of the roadway were generated.; Validation of ISENS was performed with online CCTV videos taken from three locations between 11 AM to 6 PM, and with fifteen minutes of continuous video recorded on US Highway 12. The adaptability of ISENS to different traffic conditions, different illumination conditions, and different locations was evaluated. Estimated space occupancy reasonably depicted traffic conditions of the roadway. Time headway and flow rate estimated by ISENS were statistically tested. The validation results indicated that ISENS performed robustly and accurately for traffic surveillance without camera calibration.