| dc.description.abstract | Precise vehicle trajectory information is useful for a huge array of applications in traffic science and is the cornerstone for understanding the macro-scale effects of intelligent transportation control strategies. Unfortunately, such trajectory data is difficult to obtain for every vehicle in a traffic flow, prompting a reliance on simulation or simplified testing environments for intelligent transportation technologies. This thesis seeks to address this shortage of trajectory data by providing a long-term instrument for continuous, large-scale vehicle trajectory extraction on a real roadway. This work proposes, tests, and builds the Interstate-24 Mobility Technology Interstate Observation Network (I-24 MOTION), an instrument consisting of 276 4K-resolution traffic cameras densely covering 4.2 miles of interstate roadway near Nashville, Tennessee. Novel algorithms for fast multiple object tracking utilizing sparse, well-selected crops from overall frames are proposed and implemented in both 2D and multi-camera 3D paradigms, moving towards the goal of real-time object tracking for trajectory extraction, improving algorithm speed without sacrificing accuracy. Several 3D, multiple object, multiple camera datasets are released using the system; namely a densely labeled dataset for object detector training and object tracker evaluation, and a massive 1-hour video dataset vital for assessing long-term algorithm tracking performance. The first 2 weeks of trajectory data captured from I-24 MOTION is also released, constituting the largest existing vehicle trajectory dataset both in terms of spatial and temporal length. In the near future, this instrument will enable near-effortless trajectory data extraction during open-road testing of next-generation transportation technologies proposed to make transportation safer, faster, and more efficient. | |
