|M.Sc Student||Bluminov Elena|
|Subject||Feasibility of Computerized Monitoring System for Flow and|
Risk Measurement on the Intersection
|Department||Department of Civil and Environmental Engineering||Supervisor||Professor David Mahalel|
|Full Thesis text - in Hebrew|
The Objective of this research is to examine a method that estimates the number of conflicts between vehicles in a junction, based on computer vision.
Conflicts are defined as situations that might, under certain circumstances, may result in an accident. Technically, conflict is defined as a situation when the time for an accident is shorter than a curtain threshold. The conflicts are estimated from video footage with the aid of a decoding software and classification of the situations.
This research doesn’t deal with validity of the conflicts and the their ability to actually predict accidents, but focuses on developing an ability to automatically estimate conflicts.
Collecting and analyzing information on the conflict situations is intended as a tool to estimate the risks of an accident at intersections of road. In the short term, the conflicts are meant as a substitute for the data about accidents. The use of conflicts is the fastest alternative to identify various risks which can cause accidents, and in turn assists to design tools that will help reduce the number and the amount of damage in road accidents.
One of the conflict system weaknesses used to be the field measurements, conducted by viewers. The method consumes manpower and is not accurate, because it's based on subjective judgment.
During the research a video camera was placed on an interurban junction, videotaping the course of traffic. Traffic detection was done by comparing the pixel location changes, of certain colors, between successive frames of the video. This requires isolating the changes that aren’t caused by movement of objects such as noise, lighting changes, movement of the camera itself ect., and try to capture only moving objects. A software (CarChart) was designed in order to detect motion and estimate conflicts. It was meant to read the data of the traffic in a certain junction and graphically displaying it.
Comparison of the results of the automatic decryption with direct observation of the images, detected a large number of situations in which automatic decryption requires improvement in order to increase its' reliability and to ensure a valid decoding of the frames. In addition to identifying the vehicles in various frames, there is also the issue of measuring distances of the image plane and their translation in to the earth plane.