The Problem Statement
To detect and Track Vehicles (Trucks) at the construction site, the client wanted to develop an algorithm to detect and classify the type of truck entering the construction zone. Along with tracking, the system would also log the entry and exit time of the vehicle to create a database of the entire activity and time duration for which the truck was present at the site. Considering our expertise in vision systems and years of experience in algorithm development, the client decided to associate with us to build vision-based truck recognition & tracking system for their construction sites.
Leveraging decades long experience of designing robust & high quality algorithms and vision systems, the KritiKal team analyzed each aspect of the project. To detect a moving vehicle, a background subtraction algorithm was developed using OpenCV. This module constantly detects vehicle movement in the Region of Interest (ROI). One of the major challenges was to differentiate between multiple vehicles entering the ROI zone simultaneously. Team KritiKal eliminated this challenge by assigning a unique ID to each truck. This helped in accurately tracking the activity of every truck separately.
Identifying the type of truck was another big challenge of this project. Due to the different shapes and sizes of trucks used at the construction site, it was difficult to decide on a common factor and platform to classify them accurately. Developers explored different technologies and decided to code a classification module using TensorFlow with a database of all the truck types in it.
Finally, these two modules were combined and deployed on the AWS EC2 instance to log events and update the database in real-time.
This Vision-based truck detection and classification system built by KritiKal Solutions helped in eliminating manual work at the construction site. Automatic tracking of vehicles helps in increasing the overall productivity and prevents unauthorized entry of trucks into the construction site. The client can deploy this system to any site on any camera with minimal coding efforts.