SlideShare a Scribd company logo
1 of 4
Morris Smith
Study of Sign Characteristics on I-20
January-May 2014
CEE 2699
Dr. Yichang Tsai
I. Introduction
The overall objectiveof thisresearchprojectistoassessandimprove the effectivenessof the usage
of LiDARpointcloudtechnologytoautomaticallydetectandinventorytrafficsigns. Thismethodof sign
inventory,if provensuccessful,wouldbe aconsiderable advance insignmanagementandtechnology.
Trafficsignsare a seeminglysmall yet vital componentof astable infrastructure.All of the different
typesof signsare meantto conveysimple andunderstandable messagestodriversinordertokeep
themfocusedonthe road. Thishelpstopromote safetysuchthat accidentsmaybe prevented,
especiallyonhighways,where caraccidentshave amuch higherfatalityrate thanotherroads. One way
for trafficcontrollersandengineerstostudythe efficiencyof these signsistotake an inventoryinwhich
GPS coordinatesandsigntype are considered,andthisiswhy automaticsigndetectionwouldgoa long
waytowards maintainingsustainable infrastructure.
II. Data Extraction
Before testingthe LiDARpointcloud onInterstate 20 inGeorgia,a groundtruth was neededtoserve
as the control in whichto compare itwith.Thiswas createdbymanuallyinventoryingsignsusingthe
Trimble Spatial ImagingAnalyst interface.Anapproachwastakensimilartohow a licensedtraffic
engineerwouldtake intermsof the typesof signsto be inventoried.All signswithincertainconstraints
were considered;those thatwere notwere temporarysigns,guardrail markers,electronicsigns,and
signson the right-of-wayonanentrance/exitramp.
a. Productivity
The entire setfromwhichdata was extractedcoversapproximately200 milesfromAlabamato
SouthCarolina,eastboundandwestbound.There are 136,016 framesintotal inthe set, so 400 milesof
data yieldsanaverage of 340 framespermile.The total numberof inventoriedsignsis3,847, and ittook
approximately20 hoursto collectall of them.Thisyieldsaproductivityof 3.21 signsperminute,and
113.35 framesperminute.Itishelpful toconsiderthe signdistributionalongthe highway;the highest
concentrationof trafficsignswasfoundinurbanareas and suburbanareasas well as segmentsof the
highwaythatwere underconstructionduringdatacollection.A large portionof the highway,however,
wentthroughrural areasand had sparselydistributedsigns,includingmanysectionsthatjustconsisted
of mile markersalone.
b. Pros and Cons of Data Extraction
Establishingagroundtruthdatasethad bothpositive andnegative impactsonthe overall
project.The mainreasonbehindcreatingthe groundtruthwasto provide a defaultcontrol datasetin
whichto compare LiDARresultswith.A helpful aspectof the setwasthatit had coordinatesforevery
signwithwhicha researchercouldcompare withthe correspondingsignthatwasdetectedthrough
LiDARanalysis. Thiswouldbe veryadvantageoustoanyone whowouldwanttotake a strictlyanalytical
approach to improvingthe parameters.Preliminarydataextractionwouldalsoallow new userstothe
LiDARsoftware toaccustom themselvestothe interface.
On the otherhand,the extractionof the groundtruth data was verytediousandtime-
consuming.Itisalsoimportantto note several hoursof researchwere lostduringthisphase due to
severe weatherconditions;thispushedbackthe LiDARanalysismanyweeks,andthe primaryobjective
inwhichthe projectwas aimedforwasrushedand couldonlybe carriedout ina muchshorterperiodof
time thanwhat wasdesired.
c. Issues
Three issuesinparticularwere encounteredmultipletimesduringdataextraction.The firstone
was thatoverheadsigns,mostlyroadname signs,were particularlydifficulttoaccuratelypinpointusing
the two-clicktriangulationmethod.Secondly,large vehiclessuchastrucks occasionallyblockedasign
that wason the lefthandside of the cameravan. Lastly,datapointson an opposite boundappearedto
be inaccurate.The last issue isessentiallyanobservationanddoesnotaffectthe overall project.
d. Recommendations
For future usersof thismethodof data extraction,itisrecommendedtoinventorysignsbased
on predictingwhetherornotthe LiDAR detectionwill pickitup.People whoare selectedforthiskindof
researchshouldhave anample amountof intuitiontodeduce whetherasignisbig andreflective
enoughtobe detected.Itisunderstoodthatthisprojectisfortestpurposesandwill notbe usedbyany
departmentof transportationtostudysignplacement.
III. LiDAR-Based Sign Detection
Onlyone clip(AlabamatoGeorgia,Eastbound) wasusedfortestingthe LiDARdetection.There are
five particularparametersthatwere considered whenevaluatingthe effectiveness.These are the
sensitivity,minimumelevation,the minimumdistance fromsigntosign,the maximumlateral distance,
and the minimumhitcount.Those thatwere notconsideredwere parametersthatmeasuredthe height
and widthof the signs.
a. Method of parametric study
The methodwhichwasusedto evaluate the effectivenessof the signdetectionwasabasic
visual analysis.Afterrunningthe LiDARtool,signdetectionmarkerswerecomparedwithgroundtruth
markerswithrespecttotheirlocations.The effectivenessof the parametersetwasevaluatedbasedon
whatsignsmatchedthe ground truthset,whichonesdidnot,how manymarkerswere placedateach
sign,etc.The resultswouldbe comparedforaboutone quarterof the clip,so that general assumptions
couldbe made about the rest of the data.
b. Observations
The issue that wasrun intothe most while testingthe LiDAR parameterswasthatthere were
multiple detectionmarkersatmanylarge signs.The initial conclusionwasthatthe tool was considering
the largersignsas more than one and essentiallycountedthemasmultiplesignsinextremelyclose
proximity.Inresponse,the minimumdistance fromsigntosignwasdecreasedforotherdatasets.This
decreasedthe numberof markersona decentnumberof signs.Afterwards,arealizationcame thatthe
large numberof markerswas verylikely aresultof ahighamount of reflectivityonthe signs.Afterthat
realization,the minimumhitcountwasincreased,andthe resultsshowedasignificantimprovementin
bringingthe amountof markersfor eachsignto one.
c. Recommendations
One parameterthatdealsspecificallywithreflectivenessthatcouldbe probedisthe sensitivity.
There couldbe much more analysisputintothisparameter,anditis highlysuggestedforwhomever
may pickthisprojectup to dosuch analysis.Alsorecommendedistoinclude aLiDARcamera forthe left-
handside of the road andcreate more clipsforthat; there maybe parameters thatare completely
differentthanthatof those of the right that wouldyieldoptimal performance.

More Related Content

Similar to LiDAR Analysis - Final Report

3G Drive Test Analysis for Long Call in University Campus
3G Drive Test Analysis for Long Call in University Campus3G Drive Test Analysis for Long Call in University Campus
3G Drive Test Analysis for Long Call in University Campus
ijtsrd
 
Feature Extraction and Gesture Recognition_978-81-962236-3-2.pdf
Feature Extraction and Gesture Recognition_978-81-962236-3-2.pdfFeature Extraction and Gesture Recognition_978-81-962236-3-2.pdf
Feature Extraction and Gesture Recognition_978-81-962236-3-2.pdf
TIRUMALAVASU3
 
Feature Extraction and Gesture Recognition Book.pdf
Feature Extraction and Gesture Recognition Book.pdfFeature Extraction and Gesture Recognition Book.pdf
Feature Extraction and Gesture Recognition Book.pdf
SAMREENFIZA3
 
Review Smart Traffic Management System
Review Smart Traffic Management SystemReview Smart Traffic Management System
Review Smart Traffic Management System
ijtsrd
 

Similar to LiDAR Analysis - Final Report (20)

Vocoders rare topicpptx
Vocoders rare topicpptxVocoders rare topicpptx
Vocoders rare topicpptx
 
Enhancing Traffic Prediction with Historical Data and Estimated Time of Arrival
Enhancing Traffic Prediction with Historical Data and Estimated Time of ArrivalEnhancing Traffic Prediction with Historical Data and Estimated Time of Arrival
Enhancing Traffic Prediction with Historical Data and Estimated Time of Arrival
 
IRJET- Traffic Sign Detection, Recognition and Notification System using ...
IRJET-  	  Traffic Sign Detection, Recognition and Notification System using ...IRJET-  	  Traffic Sign Detection, Recognition and Notification System using ...
IRJET- Traffic Sign Detection, Recognition and Notification System using ...
 
SMART TRAFFIC MANAGEMENT USING CLOUD
SMART TRAFFIC MANAGEMENT USING CLOUDSMART TRAFFIC MANAGEMENT USING CLOUD
SMART TRAFFIC MANAGEMENT USING CLOUD
 
Pothole Detection Using ML and DL Algorithms
Pothole Detection Using ML and DL AlgorithmsPothole Detection Using ML and DL Algorithms
Pothole Detection Using ML and DL Algorithms
 
3M Secure Transportation System.
3M Secure Transportation System.3M Secure Transportation System.
3M Secure Transportation System.
 
Automatic Detection of Unexpected Accidents Monitoring Conditions in Tunnels
Automatic Detection of Unexpected Accidents Monitoring Conditions in TunnelsAutomatic Detection of Unexpected Accidents Monitoring Conditions in Tunnels
Automatic Detection of Unexpected Accidents Monitoring Conditions in Tunnels
 
IRJET- A Survey on Reversible Watermarking Techniques for Image Security
IRJET- A Survey on Reversible Watermarking Techniques for Image SecurityIRJET- A Survey on Reversible Watermarking Techniques for Image Security
IRJET- A Survey on Reversible Watermarking Techniques for Image Security
 
3G Drive Test Analysis for Long Call in University Campus
3G Drive Test Analysis for Long Call in University Campus3G Drive Test Analysis for Long Call in University Campus
3G Drive Test Analysis for Long Call in University Campus
 
Feature Extraction and Gesture Recognition_978-81-962236-3-2.pdf
Feature Extraction and Gesture Recognition_978-81-962236-3-2.pdfFeature Extraction and Gesture Recognition_978-81-962236-3-2.pdf
Feature Extraction and Gesture Recognition_978-81-962236-3-2.pdf
 
Feature Extraction and Gesture Recognition Book.pdf
Feature Extraction and Gesture Recognition Book.pdfFeature Extraction and Gesture Recognition Book.pdf
Feature Extraction and Gesture Recognition Book.pdf
 
Drone Detection & Classification using Machine Learning
Drone Detection & Classification using Machine LearningDrone Detection & Classification using Machine Learning
Drone Detection & Classification using Machine Learning
 
DIGITAL WATERMARKING TECHNOLOGY IN INFORMATION SECURITY
DIGITAL WATERMARKING TECHNOLOGY IN INFORMATION SECURITYDIGITAL WATERMARKING TECHNOLOGY IN INFORMATION SECURITY
DIGITAL WATERMARKING TECHNOLOGY IN INFORMATION SECURITY
 
Review on IoT Based Bus Scheduling System using Wireless Sensor Network
Review on IoT Based Bus Scheduling System using Wireless Sensor NetworkReview on IoT Based Bus Scheduling System using Wireless Sensor Network
Review on IoT Based Bus Scheduling System using Wireless Sensor Network
 
A genetic based indoor positioning algorithm using Wi-Fi received signal stre...
A genetic based indoor positioning algorithm using Wi-Fi received signal stre...A genetic based indoor positioning algorithm using Wi-Fi received signal stre...
A genetic based indoor positioning algorithm using Wi-Fi received signal stre...
 
Traffic Sign Board Detection and Recognition using Convolutional Neural Netwo...
Traffic Sign Board Detection and Recognition using Convolutional Neural Netwo...Traffic Sign Board Detection and Recognition using Convolutional Neural Netwo...
Traffic Sign Board Detection and Recognition using Convolutional Neural Netwo...
 
Review Smart Traffic Management System
Review Smart Traffic Management SystemReview Smart Traffic Management System
Review Smart Traffic Management System
 
Rajshree1.pdf
Rajshree1.pdfRajshree1.pdf
Rajshree1.pdf
 
IRJET - Design Paper on Pothole Monitoring System
IRJET - Design Paper on Pothole Monitoring SystemIRJET - Design Paper on Pothole Monitoring System
IRJET - Design Paper on Pothole Monitoring System
 
Leveraging smartphone cameras
Leveraging smartphone camerasLeveraging smartphone cameras
Leveraging smartphone cameras
 

LiDAR Analysis - Final Report

  • 1. Morris Smith Study of Sign Characteristics on I-20 January-May 2014 CEE 2699 Dr. Yichang Tsai
  • 2. I. Introduction The overall objectiveof thisresearchprojectistoassessandimprove the effectivenessof the usage of LiDARpointcloudtechnologytoautomaticallydetectandinventorytrafficsigns. Thismethodof sign inventory,if provensuccessful,wouldbe aconsiderable advance insignmanagementandtechnology. Trafficsignsare a seeminglysmall yet vital componentof astable infrastructure.All of the different typesof signsare meantto conveysimple andunderstandable messagestodriversinordertokeep themfocusedonthe road. Thishelpstopromote safetysuchthat accidentsmaybe prevented, especiallyonhighways,where caraccidentshave amuch higherfatalityrate thanotherroads. One way for trafficcontrollersandengineerstostudythe efficiencyof these signsistotake an inventoryinwhich GPS coordinatesandsigntype are considered,andthisiswhy automaticsigndetectionwouldgoa long waytowards maintainingsustainable infrastructure. II. Data Extraction Before testingthe LiDARpointcloud onInterstate 20 inGeorgia,a groundtruth was neededtoserve as the control in whichto compare itwith.Thiswas createdbymanuallyinventoryingsignsusingthe Trimble Spatial ImagingAnalyst interface.Anapproachwastakensimilartohow a licensedtraffic engineerwouldtake intermsof the typesof signsto be inventoried.All signswithincertainconstraints were considered;those thatwere notwere temporarysigns,guardrail markers,electronicsigns,and signson the right-of-wayonanentrance/exitramp. a. Productivity The entire setfromwhichdata was extractedcoversapproximately200 milesfromAlabamato SouthCarolina,eastboundandwestbound.There are 136,016 framesintotal inthe set, so 400 milesof data yieldsanaverage of 340 framespermile.The total numberof inventoriedsignsis3,847, and ittook approximately20 hoursto collectall of them.Thisyieldsaproductivityof 3.21 signsperminute,and 113.35 framesperminute.Itishelpful toconsiderthe signdistributionalongthe highway;the highest concentrationof trafficsignswasfoundinurbanareas and suburbanareasas well as segmentsof the highwaythatwere underconstructionduringdatacollection.A large portionof the highway,however, wentthroughrural areasand had sparselydistributedsigns,includingmanysectionsthatjustconsisted of mile markersalone. b. Pros and Cons of Data Extraction Establishingagroundtruthdatasethad bothpositive andnegative impactsonthe overall project.The mainreasonbehindcreatingthe groundtruthwasto provide a defaultcontrol datasetin whichto compare LiDARresultswith.A helpful aspectof the setwasthatit had coordinatesforevery signwithwhicha researchercouldcompare withthe correspondingsignthatwasdetectedthrough LiDARanalysis. Thiswouldbe veryadvantageoustoanyone whowouldwanttotake a strictlyanalytical approach to improvingthe parameters.Preliminarydataextractionwouldalsoallow new userstothe LiDARsoftware toaccustom themselvestothe interface. On the otherhand,the extractionof the groundtruth data was verytediousandtime- consuming.Itisalsoimportantto note several hoursof researchwere lostduringthisphase due to severe weatherconditions;thispushedbackthe LiDARanalysismanyweeks,andthe primaryobjective inwhichthe projectwas aimedforwasrushedand couldonlybe carriedout ina muchshorterperiodof time thanwhat wasdesired.
  • 3. c. Issues Three issuesinparticularwere encounteredmultipletimesduringdataextraction.The firstone was thatoverheadsigns,mostlyroadname signs,were particularlydifficulttoaccuratelypinpointusing the two-clicktriangulationmethod.Secondly,large vehiclessuchastrucks occasionallyblockedasign that wason the lefthandside of the cameravan. Lastly,datapointson an opposite boundappearedto be inaccurate.The last issue isessentiallyanobservationanddoesnotaffectthe overall project. d. Recommendations For future usersof thismethodof data extraction,itisrecommendedtoinventorysignsbased on predictingwhetherornotthe LiDAR detectionwill pickitup.People whoare selectedforthiskindof researchshouldhave anample amountof intuitiontodeduce whetherasignisbig andreflective enoughtobe detected.Itisunderstoodthatthisprojectisfortestpurposesandwill notbe usedbyany departmentof transportationtostudysignplacement. III. LiDAR-Based Sign Detection Onlyone clip(AlabamatoGeorgia,Eastbound) wasusedfortestingthe LiDARdetection.There are five particularparametersthatwere considered whenevaluatingthe effectiveness.These are the sensitivity,minimumelevation,the minimumdistance fromsigntosign,the maximumlateral distance, and the minimumhitcount.Those thatwere notconsideredwere parametersthatmeasuredthe height and widthof the signs. a. Method of parametric study The methodwhichwasusedto evaluate the effectivenessof the signdetectionwasabasic visual analysis.Afterrunningthe LiDARtool,signdetectionmarkerswerecomparedwithgroundtruth markerswithrespecttotheirlocations.The effectivenessof the parametersetwasevaluatedbasedon whatsignsmatchedthe ground truthset,whichonesdidnot,how manymarkerswere placedateach sign,etc.The resultswouldbe comparedforaboutone quarterof the clip,so that general assumptions couldbe made about the rest of the data. b. Observations The issue that wasrun intothe most while testingthe LiDAR parameterswasthatthere were multiple detectionmarkersatmanylarge signs.The initial conclusionwasthatthe tool was considering the largersignsas more than one and essentiallycountedthemasmultiplesignsinextremelyclose proximity.Inresponse,the minimumdistance fromsigntosignwasdecreasedforotherdatasets.This decreasedthe numberof markersona decentnumberof signs.Afterwards,arealizationcame thatthe large numberof markerswas verylikely aresultof ahighamount of reflectivityonthe signs.Afterthat realization,the minimumhitcountwasincreased,andthe resultsshowedasignificantimprovementin bringingthe amountof markersfor eachsignto one. c. Recommendations One parameterthatdealsspecificallywithreflectivenessthatcouldbe probedisthe sensitivity. There couldbe much more analysisputintothisparameter,anditis highlysuggestedforwhomever may pickthisprojectup to dosuch analysis.Alsorecommendedistoinclude aLiDARcamera forthe left-
  • 4. handside of the road andcreate more clipsforthat; there maybe parameters thatare completely differentthanthatof those of the right that wouldyieldoptimal performance.