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

ParkNet: Drive-by Sensing of
Road-Side Parking Statistics
Irfan Ullah
Department of Information and
Communication Engineering Myongji
university, Yongin, South Korea
Copyright © solarlits.com
Introduction
• Mobile system that collect parking space occupancy information.
• With GPS receiver and a ultrasonic rangefinder
• Mapping
Ultrasonic sensor fitted on the side of a car detects parked cars and vacant spaces.
ROAD-SIDE PARKING CHALLENGE
• Spot reservation system
• requires exact knowledge
• requires all other vehicles to be notified
• may lead to inefficiencies
• Web based system or navigation system
• Improve traveler decisions
• Suggesting parking spaces
• adjust their prices dynamically
• Improve efficiency
Parking in slotted and unslotted areas
• Slotted: Separated by lines marked on the road
• Unslotted: without any marked parking spots
• Parking enforcement vehicle (street parking areas)
•
a sensor and connectivity to a database
Parking in slotted and unslotted areas
cont’d..
Parking in slotted and unslotted areas
• Slotted
• Easy to create occupancy map
• Parking enforcement vehicle (street parking areas)
• Unslotted: to define a space count
Distances of stretches
Number of spots
Is fixed for one parking slot (approx. 6 m)
Design goal and requirements
• SFPark project, San Francisco
• To cover 6000 parking spaces
• Detects the presence of a vehicle using a magnetometer among other sensors
• Requires repeaters and forwarding nodes on lamp posts and traffic lights.
• $250-$800 per spot, 23 million dollars
Design goal and requirements
cont’d..
• Availability of road-side parking spaces on at least an hourly basis
• A map of paid-parking spaces
• Use already installed sensors in the vehicle
• ParkNet
• Mobile sensing approach with ultrasonic sensors and GPS
Ultrasonic sensor
• Low cost than laser sensor, radar, and camera
• Reusing already present ultrasonic sensors in future vehicles
• Range and sampling rate should be large
• Time 50 ms, frequency 42 Khz, 12-255 inches every cycle
• 5Hz GPS receiver <Kernel-time, range, latitude, longitude, speed>
Prototype deployment
• Deployed on three vehicles, 2 months survey
• 57 slotted parking spots and two unslotted areas
• To obtain ground truth, we integrated a Sony Eye webcam (20fps)
• GPS trip-boxes for limiting data collection
Detection on parking spaces
Dips in the sensor reading
Sensor reading, ground truth, speed, and
output of the detection algorithm.
Challanges
• Reducing false alarms (objects other than cars)
• 20% data to train the model
• 80% to evaluate its performance
Detection algorithm
• Slotted model
•
Ultrasonic sensor
• Threshold was 89.7 inches for depth and 2.52 m for width
• Overall error rate = 12.4%.
•
GPS
• Compute the distance between the starting and ending sample
• Twice the threshold assume two cars
• Unslotted model
•
Variable space
•
Estimated against standard length (6 m)
Parked cars (blue squares), objects
other than cars (red stars)
Detection algorithm
cont’d..
• Slotted model
•
Total parking slots N, Vacant slots n, Vacant slots by sensing vehicle n',
•
Performance of the detection n'/n
•
Missed detection rate pm, false positive rate pf
• Unslotted model
•
Estimated space d', actual space between the cars is d, measure of accuracy d'/d
Detection algorithm
cont’d..
• System is 95% accurate in terms of obtaining parking counts
Slotted model
Unslotted model
Occupancy map creation
• Location coordinates provided by GPS are accurated to 3m (standard
deviation)
8 objects, 29 different runs
Correlation in the error of the
GPS location
Occupancy map creation
cont’d..
• Environmental fingerprinting approach
• Utilize ultrasonic sensor readings to correct GPS trace
• Comparing the reported location of the pattern (dips) produced by a series of
fixed objects
Using the first object in a series of 8 objects to
correct the error of the remaining 7 objects.
Occupancy map creation
cont’d..
• Slot matching
•
57 slots on a street were determined using a satellite picture from Google Earth
A comparison of the error rates in assigning
parked cars to the correct slots with and
without error correction using fingerprinting.
Mobility study
• Study of the mobility patterns of 536 taxicabs in San Francisco
busiest areas with most street-parking
Location trace of a single taxicab in San
Francisco area over a span of 30 days.
Two areas of San Francisco (i) the greater San
Francisco area, and (ii) the busiest portion
Mobility study
• Study of the mobility patterns of 536 taxicabs in San Francisco
• one can extrapolate to estimate the number of taxicabs to be fitted
with sensor
cumulative distribution function
To compute the CDF, the city is divided
into cell of size 175 m × 190 m
Cost analysis
• ParkNet
• $400 for one sensing vehicle
• $170 for PC platform: $170,
• $20 for sensor
• $100 GPS unit
• $100 for wiring and connecting components including labor
• SF park project
• cost: $120,000 (6000 fixed sensors)
Challenges faced
• DC to DC power supply
• Multilane detection through ultrasonic sensor and GPS
• Speed limitation of ultrasonic sensor
• 20 readings per second
• Speed limit in street parking areas is 35– 40 miles per hour
• Obtaining parking spot maps
• From previous data, decide unmarking and parking spots
Conclusions
• Based on over 500 miles of data collected over 2 Months
• Ultrasound sensors combined with GPS can achieve 95% accurate parking
space counts
• It can generate over 90% accurate parking occupancy maps
• Cost is reduced using mobile sensors (by a factor of 10-15)
• Using trace-driving simulation density of taxicabs can be estimated in
an urban area
• Taxicab coverage analysis will also benefit other mobile sensing
applications