Incident prediction by fuzzy image sequence analysis
Masatoshi Kimachi, Kenji Kanayama, KANAKO TERAMOTO · 2002
This study is concerned with the image sensor which is able to detect traffic incidents in a tight curve on an urban expressway. We focus on the abnormal behavior of a vehicle which caused an incident. We propose a new detection method using the image processing technique and fuzzy theory, and try to predict an incident before it happens. First, we define the "behavioral feature" which is the angle between the extracted vehicle's moving direction and the normal moving direction. We then calculate the "certainty" by fuzzy integral of the three features namely: the size, velocity and correlation value. Then we obtain the "behavioral abnormality" from the "behavioral feature" and "certainty". The "behavioral abnormality" represents the difference in behavior between the extracted car and a normally running car. Finally, a judgment is made in predicting an incident using the "behavioral abnormality" obtained from the continuous images. The proposed method is applied to some scenes in which an incident really happened and its effectiveness is verified.>