Automatic detection of traffic conditions based on analysing internet images: a data mining approach

Cameras are installed on roadside to monitor traffic. In Hong Kong, still-images of traffic conditions are captured at a fixed time interval. These images are now posted on the internet. In this research, we develop an image-based traffic monitoring approach. This approach is an important component of automatic traffic-information provision system. We analyse histograms of image's grey values. It turns out that different traffic conditions have different image's histograms. A machine-learning method is used to identify common characteristics of histograms. A prediction of traffic conditions is made using these common characteristics. Experiments on two road segments seem to support its feasibility.