Heavy-duty vehicles significantly contribute to air pollution, noise pollution, road hazards, etc., especially in urban areas. The recent advances in computer vision (CV) techniques offer local communities a promising opportunity to characterize heavy-duty traffic from traffic videos. The main advantages of the CV-based approaches lie in their non-intrusive nature, accessibility, efficiency, and affordability. However, there are no published object detection models systematically targeting truck classification. Prior related work presented methods that are often not adaptable and did not perform a comprehensive evaluation of CV-based vehicle tracking models using field measurements, undermining the understanding of their applicability in real-world settings. We developed and evaluated an integrated framework to classify different types of trucks from traffic video footage and calculate their counts, speed, and acceleration in an automated manner. The framework comprises three core components, i.e., truck detection, truck tracking, and speed estimation. Each component in the integrated framework was evaluated through field experiments to demonstrate reliability, efficiency, and accuracy. From the field experiments, we derived a recommended guideline for local communities to mount cameras to acquire reliable traffic data from the framework. We verified the feasibility, scalability, and accuracy of each step through experiments, demonstrating the potential of our approach for various applications including emission assessment and addressing other transportation-related issues for community use.