- Students Info
This project deals with information gathering and statistics about the characteristics and behavior of vehicle movement through a traffic-light-controlled intersection, for adjusting the traffic light timing to the traffic passing through the junction it is stationed in. For this purpose, a camera placed on the traffic light post and pointed at the junction is used. Thereby, an efficient, simple method of analyzing the traffic in the intersection is obtained.
In the first stage of the algorithmגs operation, a motion estimation process is applied to identify areas of the input video in which there is motion, and the directions of motion in these areas.
After that, the algorithm performs identification of the lanes in the video.
In the second stage, the algorithm analyzes the traffic load in every lane, by identifying the
vehicles in the video received from the camera, and produces load percentage for every lane as output.
The vast majority of traffic light systems today are fixed-timed and do not receive information about vehicle flow in the junctions they are governing. This form of function is problematic, since such static timing tends not to fit to the real state of traffic.
Current solutions to this problem require hardware devices which have to be installed below the road surface at great cost. The suggested system will solve the mentioned problem at the fraction of the cost, and can be integrated into a municipal traffic management system that controls traffic lights.
The project suggests a vision-based system which identifies the road surface and direction
by probing the movement of traffic on it over a period of time. It then uses this information
to distinguish between lanes and find the stop line. Using a novel algorithm to evaluate
the amount of traffic, the system produces an estimate of the traffic load at any given moment.
The sole inputs to this system are a stream of video from the camera, and the amount of lanes that are present in the road shown in said video.
The system receives a video input from a camera situated on the traffic light itself or a pole
next to it, that shows incoming traffic to the junction.
Firstly, identifying the road surface in the video is required. This is done by estimating the motion between consequent frames for a duration of a few minutes. The estimation results are then summed together to create a vector field showing the region of main movement in the video.
A background image is found by checking whether the road is empty for some time during green light.
The stop line is found by determining the closest point to the junction where vehicles stop on red light.
On this line, the centres of the lanes are marked. These points are found by dividing the stop line into the number of lanes.
From these points, “rails” representing lanes are constructed, using the vector field from the motion estimation phase.
These rails are weighted, to compensate for perspective issues resulting from the angle of the camera relative to the road.
The foreground mask is obtained by subtracting a frame from the background frame and from the previous frame.
During the load estimation process, the algorithm finds the rail parts which overlap the foreground mask. from this, a load percentage is obtained.
The system described above was built entirely in Matlab. The videos used to develop and test the various
algorithms were donated by the traffic control centre of the Municipality of Haifa. No proprietary hardware
The project shows that it is possible to gain information about traffic present in a junction, that can be
used to control it more efficiently, by solely placing a camera above it.
The entire process is completely automated, and due to the simplicity of the hardware, it is also cheap to implement.
We wish to thank the traffic control centre of the Municipality of Haifa, and in particular to Mrs Anat Gilad and Mr Tomer Niyar, for their willingness to help and donate up-to-date video files for the project.
We would also like to thank our supervisor Mr Rami Cohen, and to the staff at the Vision Laboratory – Mr Johanan Erez, Mrs Ina Talmon, and Mr Eli Appelboim, for their support and help with technicalities we were facing while implementing the project.
We are also grateful to the Ollendorff Minerva Center for its support.