Despite DC’s aims to eliminate traffic fatalities by 2024 through the Vision Zero Initiative, traffic fatalities have been on the rise during the past five years. Bike lanes are one opportunity to deliver equitable, multimodal streets for everyone by improving street safety; but bikers are incessantly put in danger by drivers who stand, park, or even drive in bike lanes. These obstacles put cyclists and other non-vehicular travelers in harm’s way as they must choose between diverting into moving traffic or risking personal injury. While nearly 5% of DC commuters travel to work via bicycle, the city consistently ranks among the most dangerous for cyclists. DC is investing to improve bike lane safety; the mayor’s recent budget proposal included $2.8 million specifically to hire additional bike lane enforcement officers and fines for obstructing bike lanes have been increased. Beyond crowd-sourcing solutions, DC has no mechanism to inform and enhance bike lane enforcement to ensure streets are safe for all travelers. To measure progress towards Vision Zero goals, DC needs the capability to monitor bike lane safety to understand patterns in bike lane obstructions as well as better tools to enforce public safety law in real-time. We propose Safe Lane, a distributed monitoring system that leverages deep learning and existing traffic camera feeds to provide bike lane enforcement with real-time notifications of bike lane obstructions as well as historical analysis to arm DC public safety agencies with the information they need to prioritize resourcing and staffing to reduce traffic fatalities.
The DC Department of Transportation has numerous programs aimed at reducing unsafe driving, including automatic speed and red light enforcement technologies. Safe Lane will provide a similar technological solution that allows public safety agencies to monitor bike lane obstructions and respond promptly. Safe Lane will leverage existing streaming data from hundreds of DDOT traffic cameras to provide the first automatic monitoring of bike lane safety anywhere. The application will primarily enhance the performance of bike lane enforcement by providing real-time notifications where and when obstructions occur. Additionally, Safe Lane’s analytics will host on-demand historical reports on bike lanes and provide new metrics for assessing bike lane safety. These dual functions will allow DC to monitor hundreds of street segments simultaneously to better qualify the problem, allocate resources, and track the efficacy of public safety and health investments.
Safe Lane is comprised of four core components: streaming video ingestion, the classification service, analytical storage and reporting, and the notification feature. Each component of Safe Lane, except for the reporting feature, can be super-charged by next-generation, gigabit technology. Safe Lane’s streaming capability ingests and processes hundreds of video streams. 5G networks would enable us to collect more durable, high-bandwidth data streams to enhance the core functionality. The core app takes video data and applies object detection models to each frame to identify obstructions in the bike lane and classify the type and duration of the obstruction. Higher frame rates and image resolution from gigabit technology would give us huge boosts in precision, accuracy, and speed. The end-to-end speed of a 5G network would also enhance the delivery of real-time notifications to end users. The application currently runs on Amazon Web Services (AWS) cloud computing products: Amazon Kinesis + Firehose, EC2, S3, Lambda, and SES. To build out the analytics feature, we will use a third-party analytics tool for interacting with the data, e.g., Tableau, Qlik, etc. Please note: For privacy purposes, Safe Lane does not store videos or images from the incoming video streams. Video is processed only to render output data qualifying the extent and type of obstruction. All downstream analytics and querying are performed on that derived data, rather than reclassifying stored images. Images would only be temporarily stored to visualize and audit classifications for testing or demonstration purposes.