Funding source: Caltrans

Contract number: 65A0674

Funding amount: $100,000

Performance period: 7/1/2023 to 6/30/2024

One of the most challenging problems in urban transportation planning is the lack of fine grain data on freight movements. Cities and regions do not know how many trucks operate in the region and have only limited information on freight flows. A particularly important information problem is the absence of a consistent and current source for freight volume and origin-destination data. Without such information, it is difficult to manage or plan for freight in metropolitan areas. This research seeks to develop a method for generating freight (truck) volume and origin-destination estimations at the traffic analysis zone level from streamed data so that estimations can be constantly updated. With our PSR-4 proposal titled "Freight Volume Modeling on Major Highway Links" we have proposed to use existing freight data sensors (e.g., WIMS, CCTV) to validate the feasibility of freight volume estimation on links on a region of interest (ROI), approximately 12 square miles in size, north of the Ports of Los Angeles and Long Beach. Specifically, we seek to validate if freight modeling is (1) feasible and (2) can be automated. Due to the COVID-19 pandemic we were only able to secure limited sensor data on the ROI. Specifically, we have collected video from Caltrans CCTV video feeds publicly available online [1], that we then used to build models to detect three truck classes (light, medium and large defined after the US DoT vehicle types nomenclature). Our preliminary results show that we can recognize those truck classes with good accuracy and therefore believe that CCTVs can be used for truck volume estimation if automated, at least during daytime with average or better visibility. We have also obtained sample TAMS [2]-[4] data which we are using to understand how accurate those sensors are with our three truck classes and then extend to other sensors such as WIMS [5] and Caltrans vehicle counting system. Finally, and most importantly we have built a simulation system that is fully configurable and utilizes real world traffic information from our ADMS system [6]. The simulation can generate truck trajectories compatible with different traffic conditions and observations for a set of modeled sensors. In parallel, we have developed algorithms to estimate (origin-destination) OD-matrices from sensor observations and we are in the process of validating their performance. This work should allow us to generate accurate origin-destination volume truck estimations by truck class and hour. Our goal with this proposal is to build on our existing work to: (1) finalize our CCTV truck detection algorithms and validate the algorithms on actual video from the area of study for different times of day to understand the performance under varying lighting conditions, (2) validate our origin-destination estimation algorithms on synthetic data under different scenarios of increasing complexity, (3) validate our origin-destination estimation algorithms on real data from the area of study and understand if existing sensor layout is sufficient to achieve good performance, and (4) use our origin-destination estimation algorithms and simulation software to model different sensor layouts and adding new sensors (to understand optimal sensor deployment and to test performance when deploying new types of sensors). If successful, other studies can leverage these tools to assess how the system can be extended beyond the area of study. Ultimately, these tools developed as part of this project would allow Caltrans decision makers to obtain freight volume information not currently available and that is needed for transportation planning. These software products can be made available widely available, e.g., by integrating into the USC METRANS/IMSC Archived Data Management System (ADMS) for the larger Los Angeles area and/or in PeMS [7] for California.