Project: #217 A Video Analytics Infrastructure Platform for Connected Vehicles and Transportation Planning Progress Report - Reporting Period Ending: March 30, 2019 Principal Investigator: Srinivasa Narasimhan Status: Active Start Date: Aug. 1, 2018 End Date: Dec. 30, 2019 Research Type: Applied Grant Type: Research Grant Program: FAST Act - Mobility National (2016 - 2022) Grant Cycle: 2018 Traffic21 Progress Report (Last Updated: March 29, 2019, 5:42 p.m.) % Project Completed to Date: 50 % Grant Award Expended: 20 % Match Expended & Document: 80 USDOT Requirements Accomplishments The first goal of this proposal is to design and develop a weatherproof video analytics system consisting of a networked edge computer and multiple cameras. This objective was completed. A cabinet was designed and assembled containing 4 CPUs, 4 GPUs, and other electronics. Working with Comcast and a contractor, our edge computing cabinet and 8 cameras were installed at the intersection of 5th Ave and Craig St. The second goal of this proposal is to develop algorithms for detecting and tracking vehicles, as well as predicting their trajectories. Data has been and is currently being collected to train deep learning models for detecting vehicles and people. Project currently has 6 Master's students (MSCV program) and 1 PhD working on it. Work is too premature to publish results. Potential to submit resulting research in top tier computer vision conferences. Impacts Deployment of platform and data collection has given rise to novel research questions that will be explored in the next 6 months. Other Collection of large amounts of a unique data set including different weather conditions at different times of day. Data being used to train a a generalized DNN model for detecting vehicles and people. Working with Zensors to obtain labeled data. Developed pipeline for easily training and evaluating models. Website built to describe project: http://www.cs.cmu.edu/~platformpgh/ Outcomes New Partners Working with company Zensors whom provides a pipeline for obtaining labeled data for training deep learning models for detecting vehicles and people. Also, provides access to data acquired from their camera network. Issues Difficulties with the stability of two installed cameras. Working with contractor to find solution. Does not affect research efforts since there are 6 operational cameras.