Capstone Spotlight: Curbside Data Analysis for Traffic Patterns
Applying the neural network model to identify trends in traffic patterns in busy city intersections
Each year, Berkeley MEng candidates embark on a two-semester capstone experience where they work with faculty or industry partners to bring solutions for real-world problems to life with engineering skills and leadership practices. Here, we place a spotlight on the capstone experience of the Curbside Data Analysis capstone team.
The team’s project objectives are to 1) understand the future of city parking involving delivery and rideshare drivers, 2) understand curb use patterns through dashcam footage, and 3) analyze curb usage trends to identify problem areas and solutions. The focus area was Bancroft Avenue in Berkeley, CA.The team partnered with UC Berkeley’s Parking & Transportation office to acquire dashcam footage from the campus’ Bear Transit Perimeter Line bus. The team is composed of Sida Li, Aravinth Paranan, Aditya Kumar, Priya Jindal (Electrical Engineering & Computer Science MEng students), Mikhail Burov (PhD student), Murat Arcak, and Alexander Kurzhanskiy (Principal Investigators).
Data Collection & Model Training
TThe team installed dashcams in the Bear Transit Perimeter Line and collected around 567 hours of dashcam video footage along the bus’s route. By labeling images of various classes of vehicles including: car, bus, FedEx, UPS, Amazon, USPS, UHaul, and other trucks; the team trained a YOLOv5 object detection neural network model. This model was then used to detect specific classes of vehicles in the collected dashcam videos and produce data visualizations and heatmaps to identify curb usage patterns.
“This project tests the waters for a potential crowdsourcing of real-time street view data collection to vehicles with dashcams. This kind of information improves safety of road users and parking efficiency by identifying free/busy curbs” says PI Alexander Kurzhanskiy.
Insights show the traffic patterns and frequencies of vehicle types over time:
What’s next?
Based on the data analysis, the team was able to collect insights on hotspots on Bancroft that seemed to have the highest amount of delivery vehicle traffic and observed patterns in the time of day that these trucks appeared most commonly. The team is also looking into how often delivery trucks seem to be on the bus lane and how this can impact the speed of the bus. Such data can be used to help identify delivery companies that may be impacting campus bus schedules, come up with a plan to better allocate parking space on the curb, and adjust parking rates dynamically to ensure maximum utilization of parking spaces and prevent double parking. This technology can also be applied to other streets or cities, allowing urban planners to gain valuable insight regarding traffic flow, parking violations, and curb allocations.
Connect with Sida Li, Aravinth Paranan, Aditya Kumar, Priya Jindal.
Edited by Ashley Villanueva
