Pavement Surface Defects Classification and Quantification over a

Pavement Surface Defects: Classification and Quantification over a Road Network Alejandro Amírola Sanz • AEPO, S. A. (Spain) • Equipment Research and Development Department • Infrastructure Management Division • aamirola@aepo. es • Portorož, Slovenia

1. Scope of the Analysis Data shown in figures only as sample of outputs 1. Image Acquisition 4. Graphics and Maps Generation 2. Surface Defects Identification 3. Quantification. Indexes Calculation Numerical Results Road Network managed by the Spanish DGC: ~30. 000 Km Portorož, Slovenia

2. What is a crack? 1. Is needed a detailed in the pavement surface with minimum “A crack is a discontinuity definition of crack for develop useful dimensions Indicators? Distresses of 1 -mm width and 25 -mm length. ” (AASHTO PP 44 -01) “A crack is assess the human pattern recognition accuracy? 2. Can we a discontinuity in the road surface that has a minimum length, can and depth. ” (PIARC automatic detection systems 3. How widthwe evaluate the. Technical Committee C 4. 2 Road/Vehicle Interaction) 60 pixels = 60 mm accuracy? Sample image: 600 pixels = 600 mm 50 pixels = 50 mm 400 pixels = 400 mm YES. Is a Crack Is it a Crack? Portorož, Slovenia

Image length: 1 m = 1000 pixels 3. Detection Methods. Automatic Systems & Human Reviewers Lane width: 4 m = 4000 pixels Portorož, Slovenia

4. Quantification Methods. Longitudinal & Surface A) Length of crack related to section length B) i. e. : 10. 6 m of crack in 1 meter lane length. C) (Severe Cracking Level) B) Portion of Surface cracked related to total surface Grid size: 10 cm x 10 cm 1 m length = 400 blocks i. e. : 108 blocks cracked in 1 meter length. 27% cracked surface Each 10 cmx 10 cm block is equivalent to 10 cm of crack length. Conversion from longitudinal to surface reference can be done Portorož, Slovenia

5. a Classification & Quantification Cracks / Distresses: 1. Longitudinal Cracking: Longitudinal Cracks Index (LCI) (IFL): 2. Transversal Cracking: Cracks Length (CL) (LF): Transversal Cracks Index (TCI) (IFT): 3. Alligator Cracking: Alligator Cracks Index (ACI) (IFM): Portorož, Slovenia

5. a Classification & Quantification Cracks / Distresses: Cracks Length (CL) (LF): Sample results (crack meter / meter section leght) 0 0. 82 4. 36 16. 7 Portorož, Slovenia

5. a Classification & Quantification Cracks / Distresses: Cracks Length (CL) (LF): CEI Cracks Equivalent Index (CEI) Conversion Range: CL (m crack/m section) CL CEI Low Medium Severe Very Severe Portorož, Slovenia

5. a Classification & Quantification Cracks / Distresses: Cracks Length (CL) (LF): Sample results (CEI) 0 0. 82 LOW MEDIUM 1. 6 2. 2 SEVERE VERY SEVERE Portorož, Slovenia

5. b Classification & Quantification Other Indexes: Sealing Index: Peeling Index: Reparation Index: Portorož, Slovenia

6. Division of the results considering wheel paths Total E D C B A Longitudinal LCIE LCID LCIC LCIB LCIA Transversal TCIE TCID TCIC TCIB TCIA Alligator ACIE ACID ACIC ACIB ACIA Total CL CLE CLD CLC CLB CLA Other indexes (sealing, …) Portorož, Slovenia

7. Accuracy levels Test Road Section: Heterogeneous 10 km section Consider the road section surface divided by a 10 cmx 10 cm grid. Measure the distresses by various trained human reviewers. Human Results: Reviewer 1 2 3 4 Average Crack 14. 5% 13. 5% 10. 9% 12. 1% 12. 8% Non Crack 85. 4% 86. 5% 89. 0% 87. 8% 87. 2% Reference Results: - 12, 8% of the blocks are crack block - 87, 2% of the blocks are non crack block - Also obtained 1 reference value per each block over all the control section The difference between average and each individual reviewer is not good enough for controlling the accuracy. i. e. : One system that provide the same values as the reference, but detects the distresses on different locations is a bad detection system. Portorož, Slovenia

7. Accuracy levels Test Road Section: Heterogeneous 10 km section “Missing Crack” Crack Real Crack marked as Crack% Real Non Crack marked as Crack% Non Crack Reviewer Result Comparison of each Human Reviewer vs. Reference Results: Comparison of the deviations block by block Real Crack marked as Non Crack% Real Non Crack marked as Non Crack% Crack Non Crack “False Positives” Target Portorož, Slovenia

7. Accuracy levels Test Road Section: Heterogeneous 10 km section Non Crack 10. 56 0. 43 2. 45 86. 56 Crack 11. 55 1. 95 Non Crack 84. 31 Reviewer 2 Crack Non Crack 1. 15 Non Crack Reviewer 1 2. 68 1. 46 85. 04 Crack Non Crack Target Crack Non Crack 11. 34 0. 85 Non Crack 0% 87. 2% non crack Missing Crack: around 1 -2. 5% False Positives: around 1 -2% Target Reviewer 4 Reviewer 3 Target 12. 8% Crack 0% Comparison of each Human Reviewer vs. Reference Results: 11. 86 Target 1. 67 86. 14 Crack Non Crack Target These values must be accepted as long as we accept that “the human pattern recognition is quite ambiguous” (PIARC Technical Committee C 4. 2 Road/Vehicle Interaction Conclusions on Evaluating the Performance of Automated Pavement Cracking Measurement Equipment) Portorož, Slovenia

7. Accuracy levels Test Road Section: Heterogeneous 10 km section Automatic Systems vs. Reference Results: Missing Crack: around 1 -2. 5% False Positives: around 1 -2% Estimated Accuracy level obtained by human reviewers At the moment, we are using automatic detection system that have: ~3% Missing Crack ~6% False Positives Is reasonable expect to develop automatic systems that can recognize distresses better than human reviewers? Can the automatic detection systems get better accuracy levels that humans? Portorož, Slovenia

8. Conclusions A detailed definition of what is a crack is not a critical issue for the study and characterization of pavement distresses at the Network Level. Definition of indexes is critical to obtain good and useful results. These indexes can be customized for the end user (Road Administrators) Human accuracy levels should be considered as a reference when automatic detection systems are being developed. Portorož, Slovenia