ParkNet

1
ParkNet
Drive-by Sensing of Road-Side Parking Statistics

• Sutha Mathur, Tong Jin, Nikhil Kasturirangan,
• Janani Chandrashekharan, Wenzhi Xue, Marco
  Gruteser, Wade Trappe
• Rutgers University

Michael Betancourt UCF - EEL 6788 Dr. Turgut
2
Overview

1. Introduction
2. Design Goals and Requirements
3. Prototype Development
4. Parking Space Detection
5. Occupancy Map
6. Mobility Study
7. Improvements
8. Conclusion

3
Introduction - Problems

• Traffic congestion costs tons of money
• 4.2 billion lost hours
• 2.9 billion gallons of gasoline wasted
• Looking for parking contributes to these numbers
• Lack of information
• Hard to determine best prices for meters and
  where they should be placed
• Current parking detection systems are costly

4
Introduction - ParkNet

• Drive-by Parking Monitoring
•  Uses ultrasonic sensor attached to the side of
  cars
• Detects parked cars and vacant spaces
• Attaches to vehicles that comb through a city
  (taxi, police, etc.)
• Location accuracy based on GPS and environmental
  fingerprinting

5
Introduction - Objectives

• Demonstrating  the feasibility of the mobile
  sensing approach including the design,
  implementation and evaluation of the system
• Proposing and evaluating a method of
  environmental fingerprinting to increase location
  accuracies
• Showing that if the mobility system were
  currently attached to operating taxis, it would
  operate with enough samples to determine parking
  availability

6
Design Goals - Real-time Information

• Improve traveler decisions with respect to mode
  of transportation
• Suggesting parking spaces to users driving on the
  road
• Allow parking garages to adjust their prices
  dynamically according to demmand
• Improve efficiency of parking enforcement in
  systems that utilize single pay stations for
  multiple parking spots

7
Design Goals - Parking Information

• Space count
• Sufficient for most parking applications
• Occupancy Map
• Useful for parking enforcemen

8
Design Goals - Cost and Participation

• Low-cost Sensors
• Typical per spot parking management systems
  ranges from 250 to 800 per spot
• Current systems are difficult to place in areas
  without marked parking spots
•  Low Vehicle Participation
• Be able to function without a lot of cars fitted
• Keep costs down

9
Prototype Development - Hardware

• Moxbotix WR1 rangefinder
• Waterproof
• Emits every 50ms
• 12-255 inches 
• PS3 Eye webcam
• 20 fps
• Used for ground truth
• Not in production
• Garmin GPS
• Readings come at 5Hz
• Errors can be less than 3m

• On-board PC
• 1GHz CPU
• 512 MB Ram
• 20 GB HD
• PCI WiFi card
• 6 USB ports

10
Prototype Development - Deployment

• System was placed on 3 vehicles
• 3 specific areas were marked off to be analyzed
• Data was collected over a 2 month period
• Drivers were oblivious to the data collection
• All range sensor data is tagged
  withKernel-time, range, latitude, longitude,
  speed

11
Prototype Development - Verification

• PS3 Eye
• Mounted just above the rangefinder
• Took pictures at 20fps that were time tagged
• Each picture was manually checked to see if there
  was a car parked
• This was used to verify the data collected from
  the system

12
Parking Space Detection - Challenges

• Ultrasonic sensor does not have a perfectly
  narrow-width
• GPS Errors
• False alarms
• Other impeding objects Trees, people, recycling
  bins
• Missed detections
• Parked vehicles classified to be something other
  than a parked car

13
Parking Space Detection - Dips

• A "dip" is a change in the rangefinder readings
  which usually occurs when there is an object in
  view

Two Cars Parked Together
Far
Close
14
Parking Space Detection - Algorithms

• Slotted Model
• Determines which dips are classified as cars
• Subtracts the total number of cars found with the
  total number of spaces available in the area
• Unslotted Model
• Determines which dips are classified as cars
• Measures the distance between dips to see if it
  is large enough to fit a car
• Training
• 20 of the data is used for training
• 80 of the data is used for evaluating performance

15
Parking Space Detection - Slotted

• Slotted Model Accuracy

16
Parking Space Detection - Unslotted

• Unslotted Model Accuracy

17
Occupancy Map - GPS Error

• Selected 8 objects and determined their absolute
  GPS position using Google Maps
• Corresponded the GPS reading gathered from the
  trials to the objects
• Used the reading from one object to correct the
  others

18
Occupancy Map - Environmental Fingerprinting

• ?Fixed objects in the environment used to
  increase positional accuracy
• Recognition Walkthrough
• GPS coordinates indicate system is near known
  object
• Parses rangefinder readings
• Determines what is not a parked car
• Tries match the pattern with the known object
• If object found, correct position if within 100m

19
Mobility Study - Taxicab Routes

• Public dataset of 536 taxicabs GPS position every
  60 seconds
• Routes were approximated by linear interpolation
• Found that taxicabs spend the most time in
  downtown areas where parking is scarce
• Determined the mean time between cabs visiting a
  particular street.

20
Mobility Study - Taxicab Mean Time
Greater San Francisco
Downtown San Francisco
21
Mobility Study - Cost Analysis

• Current Cost
• Parknet (400 per sensing vehicle) x (number of
  vehicles needed to get desired rate of detection)
• Fixed Sensor (250-800 per space) x (number of
  spaces)
• Uses opportunistic WiFi connections to transfer
  data
• Easily managed due to the much smaller number of
  fixed sensors
• Example
• 6000 parking spots
• Parknet 300 cabs, 80 coverage every 25 minutes,
  0.12 million
• Fixed Sensor 1.5 million

22
Improvements

• Multilane Roads
• Moving cars could be determined by long dips
• Rangefinder would need to be longer
• Speed Limitations
• Sensors currently work best at speeds below 40mph
• Obtaining Parking Spot Maps
• Difficult for large areas
• Algorithms could determine location surroundings
  after data collection has been started
• Using vehicles current proximity sensors

23
Conclusion

• Data collected
• 500 miles over 2 months
• Accuracy
• 95 accurate parking space counts 
• 90 accurate parking occupancy maps
• Frequency and Coverage
• 536 vehicles equipped
• Covers 85 every 25 minutes of a downtown area
• Covers 80 every 10 minutes of a downtown area
• Cost Benefits
• Estimated factor of 10-15 times cheaper than
  current systems
• Questions?

24
Links

• Fixed Parking System (SFpark)
• http//sfpark.org
• http//vimeo.com/13867453