Friday, 20 July 2012

Congestion? Not a problem!

This post is about another course project which I did for my Master's degree in Computer Vision Engineering. It's as simple and innovative as it could get. The idea is to estimate the flow of traffic in real-time using the CCTV video. Four categories of congestion/flow are first categorized, which are Low, Medium, Medium-High and High Congestion.

Here is a short demo video for this project:

While implementing this, the most difficult part was to distinguish between Medium-High and High Congestion, as both types of congestion have data which overlaps and is difficult to model with a machine learning algorithm such as K-means Clustering.

To achieve accurate results, I implemented models for both background and traffic. The main aim of background model is to extract the background and then use it to track the motion of the cars. The number and motion of the cars is then used together in the traffic model to extract the flow.

This project was initially implemented on BeagleBoard to utilize its portable application.

Typical applications for this project are in low-cost traffic congestion control systems. Multiple BeagleBoards can be used across different roads, throughout a city. Utilizing the networking capability of the BeableBoard, these devices can form a networked congestion control system, which will not require a centralized control server/office.