Counting bicycles and pedestrians is becoming essential in evaluating the different modes of transportation and providing better service to the users of dedicated bicycle paths. Accurate counts can lead to safer and more effective designs as well. Manual counts are not accurate, and during this period of reduced agencies' budgets may not be plausible, especially when one is interested in bicycle usage in complex urban environments. This project deploys the available technologies and creates a portable system that fuses the measurements from different sensors. The methods used include semi-supervised learning in conjunction with dictionaries of object models as the theoretical underpinnings. This research builds on earlier work in the field where the different available technologies for MnDOT were scoped. Counts from different locations (with consultation with MnDOT personnel) are delivered utilizing a system based on multiple cameras and sensors such as tube-based devices. A hardware prototype is in development.