Project: Novel ML-Framework For Reliability Monitoring Systems

Travel Time Reliability Monitoring Systems

Keywords: ML framework, adaptive filtering, sensor fusion, pattern recognition

Research Overview
This seminal project focused on developing an end-to-end ML framework for travel time reliability monitoring systems.
Products

  • Guidebook for state and federal agencies on how to establish and monitor such systems
  • Algorithmic framework for heterogenous sensor fusion
  • New algorithms for computing joint density functions
  • Algorithms for dynamic route guidance systems
Technical Contribution 1: Timestamp Attribution Algorithm
  • Ascribing accurate timestamps from multiple pings is a very critical issue in low-level data processing
  • I developed a low-level filtering algorithm that analyzes raw data and extracts the timestamp associated with a given location
Technical Contribution 2: Adaptive Filtering Algorithm to Identify Incidents
  • Raw sensor data is very noisy
  • My real-time adaptive filtering algorithm extracts a signal from the noise
  • It enables the identification of “anomalies” in the system
Technical Contribution 3: Sensor Fusion Algorithms
  • Heterogenous sensor data sources
  • System detectors, cameras, Bluetooth, GPS, etc.
  • Each sensor mode provides different pieces of information
  • I developed multi-sensor fusion algorithms to combine data streams
Technical Contribution 4: Classification Algorithms to Characterize Reliability
  • System reliability is influenced by factors such as weather, incidents, special events etc.
  • I developed a classification algorithm that segments data into appropriate regimes
Technical Contribution 5: Template Matching Algorithms
  • The popular KS-test didn’t work well for assessing the similarity between probability density functions
  • I developed an alternate template-matching algorithm to overcome this hurdle
Technical Contribution 6: Route Synthesis Algorithm
  • The random variables demonstrated have strong dependencies.
  • Conventional ML-algorithms can’t be used for this synthesis
  • I developed a numerical integration algorithm that takes these statistical dependencies into account and accurately synthesizes the route-level PDFs
Notes
  • In 2011, I developed these ideas and wrote a 132-page use-case document targeted towards practitioners.
  • Many of these same ideas are now implemented as features in Google Maps.

Get in Touch