Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Visual Exploration of Trajectory Data

159 views

Published on

Nikola Markovic, University of Utah; Mark Franz, University of Maryland CATT Lab

Published in: Education
  • Be the first to comment

  • Be the first to like this

Visual Exploration of Trajectory Data

  1. 1. Visualization of Processed Trajectory Data Seth Miller – University of Utah https://markoviclab.civil.utah.edu/
  2. 2. INRIX Trajectories: 2.3 Million Trips
  3. 3. Capturing 3% of Trips Crossing UDOT Sensor Stations
  4. 4. Waypoint Heatmap
  5. 5. Salt Lake City Inflow and Outflow
  6. 6. With a Whole Population of Trips  Calculate AADT (Average Annual Daily Traffic)  Estimate vehicle delays varied spatiotemporally  Measure impacts of special events on traffic patterns  Use for Map-21  Improve transportation planning outcomes (Utah 30-year plan)
  7. 7. From Sample to Population: Just Add a Column to File from INRIX  We compare to baseline model from MIT – ICWG (Iqbal, Chodhurry, Wang, and Gonzalez) Count of a Single INRIX Trip Estimated Count of Trips Just Like this One A Trip’s Scaling Factor Origin Destination Noon to 1 pm 1pm to 2pm Salt Lake Weber 32 31 Salt Lake Davis 31 30 Salt Lake Utah 30 29 Trip ID Origin County Destination County Departure Time Trip Count Example Salt Lake Weber 1:25 pm 31
  8. 8. 3. Control accuracy vs overfitting Model Overview 2. Decrease overfitting 𝑒 𝑘𝜏 = ෍ 𝑝∈𝑃 ෍ 𝜏∈𝑇 𝑐 𝑝 𝑡𝑘𝜏 ∗ 𝑠 𝑝 𝑡 min 𝑣 𝑝 𝑡 1. Make estimated counts close to ground truth 5. Tie scaling factors to systemwide penetration rate 𝑠 𝑝 𝑡 = 1 𝑟 𝑡 + 𝑣 𝑝 𝑡 s. t. 4. Estimate counts by scaling trajectories ෍ 𝑘∈𝐾 ෍ 𝜏∈𝑇 𝑒 𝑘,𝜏 − 𝑔 𝑘,𝜏 + 𝛾 ∗ ෍ 𝑝∈𝑃 ෍ 𝑡∈𝑇 𝑣 𝑝 𝑡 2
  9. 9. Test on Stations Unseen For Training 0% 10% 20% 30% 40% 50% 60% 0 1 2 MedianAbsolute PercentError γ (Higher γ = Higher Overfitting Penalty) Tuning Overfitting Parameter on Cross Validation Set Cross Validation Training
  10. 10. Results Comparison 0% 20% 40% 60% 80% 100% 0% 50% 100% CumulativePercent Absolute Percent Error CDF of Error on Training Data Baseline: 37% Proposed: 11% 0% 20% 40% 60% 80% 100% 0% 50% 100% CumulativePercentage Absolute Percent Error CDF of Error on Testing Data Basline: 32% Proposed: 18% Baseline Proposed
  11. 11. Origin Destination Counts
  12. 12. Looking to the Future: Other States and Other Granularity
  13. 13. Acknowledgement
  14. 14. Center for Advanced Transportation Technology Laboratory – Mark Franz

×