Remote Sensing for Asset Management Shauna Hallmark Kamesh

Remote Sensing for Asset Management Shauna Hallmark Kamesh Mantravadi David Veneziano Reginald Souleyrette September 23, 2001 Madison, WI

The Problem/Opportunity • DOT use of spatial data – – Planning Infrastructure Management Traffic engineering Safety, many others • Inventory of large systems costly – e. g. , 110, 000 miles of road in Iowa

The Problem/Opportunity • Current Inventory Collection Methods – Labor intensive – Time consuming – Disruptive – Dangerous

Data Collection Methodologies • Manual (advantages/disadvantages) • • • low cost visual inspection of road accurate distance measurement workers may be located on-road difficult to collect spatial (x, y) • Video-log/photolog vans (advantages/disadvantages) • rapid data collection • digital storage • difficult to collect spatial (x, y)

Data Collection Methodologies • GPS (advantages/disadvantages) • highly accurate (x, y, z) • can record elevation • time consuming if high accuracy is required • workers may be located onroad

Data Collection Methodologies • Remote sensing (advantages/disadvantages) • Data collectors not located on-site • Initially costly but multiple uses • Can go back to the images

Research Objective • Can remote sensing be used to collect infrastructure inventory elements? • What accuracy is possible/necessary?

Remote Sensing • "the science of deriving information about an object from measurements made at a distance from the object without making actual contact” Campbell, J. Introduction to Remote Sensing, Second Edition. • Applications in many fields such as forestry, Oceanography, Transportation

Remote Sensing • 3 types 1) space based or satellite • Images acquired from space 2) airplane based or aerial • Images acquired form aerial platforms like high, low altitude airplanes and balloons. (USGS) 3) in-situ or video/magnetic

Research Approach • Identify common inventory features • Identify existing data collection methods • Use aerial photos to extract inventory features • Performance measures • Define resolution requirements • Recommendations

Application • Use of Remote sensing to collect features for the Iowa DOT’s Linear Referencing System (LRS) • Datum – Anchor points – Anchor sections • Business data – Inventory features

Datum • Anchor points – Physical entity – (X, Y) – Intersection of 2 roadways – Intersection of RR and roadway – Edge of median – Bridges • Anchor sections – Measurement of distance between anchor points along roadway Anchor point Anchor section

Datum Accuracy Requirements Ø Anchor points Ø ± 1. 0 meter Ø Anchors sections Ø ± 2. 1 meter

Common Business Data Items • HPMS requirements • Additional Iowa DOT elements § Section Length § Number of Through Lanes § Surface/Pavement Type § Lane Width § Access Control § Median Type § Median Width § Parking § Shoulder Type § Shoulder Width - Right and Left § Number of Right/Left Turn Lanes § Number of Signalized Intersections § Number of Stop controlled Intersections § Number of Other Intersections

Imagery Datasets • • 2 -inch dataset - Georeferenced 6 -inch dataset - Orthorectified 2 -foot dataset – Orthorectified 1 -meter dataset – Orthorectified – simulated 1 -m Ikonos Satellite Imagery * not collected concurrently

Performance Measures • Establishing geographic location of anchor points and business data – Positional accuracy – Variation between operators for locating elements (Operator Variability) – Ability to recognize features in imagery (Feature Recognition) • Calculation of anchor section lengths • Establishing roadway centerline

Positional Accuracy • Root Mean Square (RMS) • Imagery position vs. position w/ GPS (centimeter SE corner of intersecting sidewalks horizontal accuracy) • 2 easily identified features selected – Could be identified in all 4 datasets – Had a distinct point to locate SE corner of drainage structure

Positional Accuracy • 2 -inch, 6 -inch, 24 -inch met accuracy requirements of Iowa DOT LRS for anchor points • Even for 1 -meter RMS < 2 meters • 95% of points were located within < 3. 5 meters for all datasets --- sufficient accuracy for most asset management applications

Operator Variability • For manual location of features • How much of spatial error can be attributed to differences in how data collectors locate objects Variation among observers in spatially locating a point

Operator Variability • 7 operators located 8 sets of features – – – – Edge of drainage structure as located by 7 operators Traffic signal posts Drainage structures Pedestrian crossings Center of intersections Center of driveways RR crossings Bridges Medians • Specific instructions for locating (i. e. SE corner of bridge) • Compared variability among observers

Operator Variability (results) • Only 3 features could be identified consistently in all 4 datasets – Driveways --- RR Crossings – Center of intersections • 5 other features identified in 6 -inch & 2 inch datasets

Operator Variability (results) • Certain features, such as railroad crossings, could be located with less variation than features such as driveway centers (less distinct) • mean variability < 0. 5 meters – Drainage structures, driveways, traffic signal posts, pedestrian crossings (2 and 6 -inch tested only) • mean variability >= 0. 5 m & < 1. 0 m – Medians (2 & 6 -inch tested only, RR crossings) • mean variability >= 1. 0 m – Intersections, bridges • Significant variability in features used as anchor points • Variability ~ allowed error (1. 0 meter)

Feature Identification • Points can be located within allowance for anchor points (± 1. 0 m) for all but 1 -meter • Even 1 -meter rms < 2. 0 meters, sufficient for most asset-related applications • But can features be consistently recognized IP (%) = (Fa/Fg) * 100 • % of features recognized in imagery compared to ground count Extraction of features from 6 -inch image

Feature Identification

Feature Identification • Of 21 features – 2 -inch: 100% identified consistently – 6 -inch: > 80% identified consistently • Signs, median type, stopbars, utility poles – 24 -inch: < 50% consistently identified • 6 features not identified at all – 1 -meter: < 25% consistently identified • 8 features not identified at all

Calculation of anchor section lengths • Linear measure along roadway centerline between anchor points • Iowa DOT LRS requires ± 2. 1 m • Established centerline and measured for 7 test anchor section test segments • Compared against DMI values from Iowa DOT LRS Pilot Study • Also collected distance using Roadware DMI van collected at ± 10 m) (but

Anchor Section Results • None of the methods met ± 2. 1 m RMS required for anchor section distances **** Iowa DOT study found 6 -inch met accuracy requirement *** • All imagery: RMS < 8 meters • All imagery: mean < 2 m

Establishing Roadway Centerline Typical Segment on Dakota (imagery and DGPS) Deviation from datum (m) • Compared centerline representation of 3 methods – Imagery – Video. Log DGPS – Roadway DGPS

Establishing Roadway Centerline Worst Alignment on Union (DGPS) Deviation from datum (m)

DGPS Traces from Iowa DOT LRS Pilot Study Nevada, IA

Conclusions • Most significant issue with imagery – At lower resolutions, difficult to identify features • Spatial accuracy for all imagery datasets comparable • Limiting factor is ability to consistently identify features • Minimum of 6 -inch required for identification of features • 1 -meter or 24 -inch: – for measurement of centerline – Identification of large features

Questions?