Camera based traffic counting is a frequently used technology in smart city applications for estimating the traffic load of the road network. Many solutions use a fixed camera and image processing techniques. As image processing is relative resource consuming, this solution includes a powerful embedded system to process the video images in real-time, so that the communication traffic is limited to the transmission of retrieved statistics. %The embedded system limits the number of applicable methods.
The motion detection systems often use methods based on color changes so shadows are usually classified as moving objects. The aim of this thesis is to examine the effect of sharp shadows caused by strong sunlight and give solution to the misclassification problem.
In my thesis, I introduce the traffic counting system and give a detailed specification of how the system detects and counts vehicles.
After that I show a simple method to the problem which uses the information about the lanes. Then I present my general shadow removal module. Next I describe existing shadow removal methods and the way I implemented them.
Finally I evaluate the result of the implemented methods, summarize the experiences and mention a few directions for future work.