Using Online CCTV Image Sequences For Real-time Traffic Estimation

As an important component of Intelligent Transportation Systems (ITS), the Advanced Traveller Information System (ATIS) disseminates real-time traffic information to the public so as to alleviate the traffic congestions. To achieve this goal, the accurate estimation of traffic flow parameters in real-time is the key issue for the implementation of ATIS.In order to identify the traffic condition timely, and accurately, a large amount of traffic detectors should be set up to cover the whole network. However, generally it is impossible to install sufficient traffic detectors in the network due to limited budget. In reality, the number of installed traffic detectors is much less than the requirement of ATIS applications. On the other hand, there exists a considerable amount of Closed Circuit Television (CCTV) cameras in the network for the purpose of traffic surveillance and monitoring. These CCTV cameras could provide a sequence of images of real-time traffic conditions in a certain time intervals (say 1-3 minutes). Surprisingly, the image sequences generated by installed CCTV cameras have not been used for traffic flow parameters estimation. This may be due to the low update rate of CCTV images and the possible movements of CCTV camera (such as zoom in/out, tilt, and pan) which make these CCTV images very difficult to process by using existing video-based methods.In view of this, a new image processing method is proposed to estimate traffic flow parameters by using these CCTV image sequences with a low update rate. The proposed method has significant theoretical and practical advances as it could achieve accurate traffic flow parameter estimation without the requirement of any modification to existing CCTV camera. The main contributions of this study are summarized as follows:(1) The characteristics of CCTV image sequences, such as low update rate, low signal-to-noise, low resolution and camera's movable features, are analyzed in detail. The difference between estimating traffic flow parameters using video and CCTV image sequences also be discussed in this paper..(2) A new image processing method is proposed to estimate traffic density and travel speed based on these CCTV image sequences with a low update rate. The proposed method consists of four major components, including an image pre-processing module, a background modeling module, a vehicle extraction module and a traffic estimation module. (3) A novel background modeling algorithm is developed to rebuild the background from low updated CCTV images. The developed background modeling algorithm could effectively rebuild background using only a few CCTV images (say less than 10 frames), and update the background when the environmental condition changes. The experimental results showed that our method could efficiently and robustly rebuild background at various traffic conditions..(4) A new approach is introduced to estimate traffic density by using the rate of vehicle occupied area to road area, and the traffic density is convert to traffic speed by using the calibrated speed-density relationships. The numerical result conducted in a real-world system showed that, the proposed approach could accurately estimate traffic density and traffic speed from the CCTV image sequences with a low update rate.