Contract number: 69A3551747109

Funding amount: $99,980

Performance period: 08/16/20 to 08/15/21

Link to full seminar video: https://youtu.be/KIEom3H6ha8

With this research we would like to validate the feasibility of freight volume estimation on major highways links from accurate but sparse sensor data, by studying a restricted control area covering approximately 12 square miles around the Ports of Los Angeles and Long Beach where freight volume is most relevant. We researched creating a real-world dataset of real-world data in this region, however, due to the COVID 19 pandemic we had to limit the collection to publicly available Caltrans CCTV video footage and leveraged our ADMS traffic database to generate synthetic data. We have implemented state-of-the-art computer vision algorithms, which were used to classify trucks and define truck categories optimized for best performance. These results show that it is feasible to use CCTV cameras to detect and classify trucks and that the process can be fully automated. In parallel, we created a truck simulator to generate realistic truck trajectories between predefined locations, and developed algorithms to estimate freight on links with different heuristics. Our approach provides the best results by relating compatible observations across sensors using travel-time information estimated on current traffic conditions. Our preliminary results show that freight volume estimation on major highways links is feasible.