A cutting-edge platform that repurposes existing hardware—from agency, city, and government cameras—to deliver real-time public transit analytics. Our solution leverages privacy-preserving object detection to count passengers (without recording personal identities) and processes these counts via a distributed pipeline. Imagine a city like Edmonton tracking bus ridership across 900+ buses in real time to optimize transit operations and resource allocation.
Youtube Demo: https://youtu.be/CI5sKFSr3U4
Smart Transit Real-Time Analytics transforms existing camera infrastructure into a powerful tool for modern urban management. By implementing state-of-the-art machine learning models for object detection (focused solely on counting objects to maintain privacy), we provide actionable insights through a fully distributed, real-time pipeline. Our system empowers cities and transit agencies to:
- Optimize Resources: Deploy more buses in high-demand areas.
- Improve Traffic Strategies: Develop data-driven strategies for smoother traffic flow.
- Enhance Emergency Response: Monitor passenger volumes to trigger timely alerts and updates.
- Real-Time Data Pipeline:
- Ingests live video feeds from cameras installed on buses, trains, or public spaces.
- Leverages OpenCV to simulate frame transmission to Kafka Topics.
- Privacy-Preserving Object Detection:
- Only counts individuals without storing or transmitting personal data.
- Detect and estimate transit occupancy using a YOLOv8 ML object detection model.
- Distributed Processing:
- Utilizes Confluent Apache Kafka for data streaming.
- Processes streams with Confluent Apache Flink for real-time analytics.
- Scalable Dashboard:
- Feeds real-time insights to dynamic dashboards, enabling instant visualization.
- Uses Django, PostgreSQL, and HTML & CSS to power our APIs, databases, and front-end dashboard.
- Containerized Deployment:
- Uses Docker Compose to containerize and simplify our deployment.
Diagram (Conceptual):
flowchart LR
A[Camera Feeds] --> B[Apache Kafka]
B --> C[Apache Flink]
C --> D[ML Object Detection]
D --> E[PostgreSQL DB]
E --> F[Dashboard & Analytics]
-
Camera Feeds:
Existing hardware captures video streams from transit vehicles or public areas. -
Data Ingestion via Apache Kafka:
Streams video data into a Kafka cluster, ensuring high-throughput and fault-tolerant delivery. -
Real-Time Processing with Apache Flink:
Processes data in real time to run ML object detection models, extracting counts from video frames. -
Dashboard Integration:
Processed data is fed into dashboards to display live transit statistics and insights.
- Confluent Apache Kafka
- Confluent Apache Flink
- ML frameworks: YOLOv8
- Computer vision libraries: OpenCV
- Docker
- Real-time Django Dashboards
- PostgreSQL databases for efficient data storage
Our platform offers a broad range of benefits that extend well beyond simple transit monitoring:
-
Extended Hardware Lifespan & Maximized ROI:
Use existing camera infrastructure to extract new value, maximizing the return on previous investments and reducing the need for new capital expenditures. -
Operational Efficiency:
Real-time insights empower transit agencies to adjust schedules and routes dynamically, reducing wait times and ensuring optimal resource allocation. -
Resource Optimization:
Identify high-demand areas and peak usage times to strategically allocate buses, trains, and other transit assets, thereby improving service efficiency and reducing operational costs. -
Enhanced Traffic Management:
Utilize real-time data to develop dynamic traffic strategies, alleviate congestion, and improve overall urban mobility. -
Data-Driven Policy Making:
Provide city planners and decision-makers with actionable, real-time data to inform long-term transit strategies and urban planning initiatives. -
Enhanced Public Safety and User Experience:
Real-time monitoring ensures rapid decision-making during emergencies and enhances overall rider satisfaction through improved transit efficiency. -
Seamless Integration with Smart City Initiatives:
Integrate with broader smart city frameworks, supporting initiatives like adaptive traffic control, smart parking, and comprehensive urban planning dashboards. -
Customizable Analytics and Reporting:
Offer flexible analytics and reporting tools that can be tailored to the specific needs of different agencies, ensuring that insights are both relevant and actionable.
-
Ingesting Data:
Direct camera feeds are ingested through the Kafka pipeline. -
Processing:
Apache Flink runs ML object detection models in real time, producing count data.
To further enhance real-time analytics and improve decision-making, our platform incorporates several optimization techniques:
- Instead of relying solely on periodic API polling, the frontend subscribes to an Apache Kafka topic via a WebSocket connection.
- This ensures that updates on passenger counts, bus occupancy, and traffic conditions are instantly pushed to the dashboard, reducing latency and improving responsiveness.
- To provide visual verification without compromising privacy, the system can display blurred, anonymized images next to live dashboard analytics.
- This feature allows transit agencies to validate passenger counts and identify potential operational issues, such as overcrowding or improper seating distribution.
- The anonymization ensures compliance with data privacy laws while enhancing situational awareness.
- GPS integration allows for real-time tracking of each bus, displaying its exact location on a city map.
- Overlaying real-time passenger load on each tracked vehicle enables better routing and adaptive scheduling, ensuring optimized resource allocation.
- Instead of treating the entire bus as a single occupancy metric, our system applies spatial fragmentation by segmenting the vehicle into multiple sections (e.g., front, middle, back).
- Each section’s occupancy is analyzed separately, allowing for:
- Identification of underutilized spaces within the bus.
- Adjustments to seating arrangements for better passenger distribution.
- Recommendations for dynamic bus design optimizations based on usage patterns.
- The insights generated from spatial fragmentation, real-time occupancy, and traffic flow data provide governments with the ability to:
- Develop better public transportation policies.
- Implement targeted infrastructure improvements.
- Justify budget allocations for fleet expansion or reallocation.
- Introduce incentive programs to distribute passenger load across different hours or routes.