FastCrack: Real-Time Pavement Crack Segmentation

Zhuang Yue; Xiaodong Chen; Yi Wang; Huaiyu Cai; Weixi Yan; Liying Hou

doi:10.3788/LOP220754

Laser & Optoelectronics Progress, Volume. 60, Issue 6, 0615006(2023)

FastCrack: Real-Time Pavement Crack Segmentation

Zhuang Yue¹, Xiaodong Chen^1、*, Yi Wang¹, Huaiyu Cai¹, Weixi Yan², and Liying Hou³

¹Key Laboratory of Photoelectric Information, Ministry of Education, School of Precision Instrument and Opto-Electronics Engineering, Tianjin University, Tianjin 300072, China

²Technology and Information Center, Tianjin Expressway Group Incorporated Company, Tianjin 300384, China

³Maintenance Department, China Roads Engineering Consulting Incorporated Company, Shijiazhuang 050030, Hebei, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Architecture hyperparameter selection framework

Download full size

Fig. 2. Width, depth, and downsampling rate experiment of the model

Download full size

Fig. 3. Convolution block type selection

Download full size

Fig. 4. Convolution type search algorithm

Download full size

Fig. 5. Comparison result of parameter indicators of each model, D represents model depth and R represents downsampling rate.

Download full size

Fig. 6. Hyperparameters selected by grid search

Download full size

Fig. 7. Network structure of FastCrack-SPOS

Download full size

Fig. 8. Crack segmentation results

Download full size

Table 1. $Δ R_{I o U}$ and $Δ S_{F P S}$ of each model relative to baseline model

View table

Table 1. $Δ R_{I o U}$ and $Δ S_{F P S}$ of each model relative to baseline model

w	D	R	IoU /%	$S_{F P S}$ /（frame·s^-1）	$Δ R_{I o U}$ /%	$Δ S_{F P S}$ /%	$\frac{Δ R_{I o U}}{Δ S_{F P S}}$
80	D3	1/8	63.40	32	3.84	322	0.01193
64	D3	1/8	63.67	59	4.11	295	0.01393
48	D3	1/8	63.15	78	3.59	276	0.01301
32	D3	1/8	62.83	152	3.27	202	0.01619
16	D3	1/8	61.51	231	1.95	123	0.01585
8	D3	1/8	59.56	354		322

Table 2. Validity analysis of FastCrack-SPOS model

View table

Table 2. Validity analysis of FastCrack-SPOS model

Method	IoU /%	Precision /%	Recall /%	F1 /%	FLOPs /10⁹	$S_{F P S}$ /（frame·s^-1）	Params /10⁶
FastCrack-Basic	62.83	75.42	79.00	77.17	12.64	152	0.61558
FastCrack-Fac	61.64	74.91	77.67	76.27	9.19	163	0.438
FastCrack-Sep	61.26	74.77	77.22	75.97	4.42	125	0.208
FastCrack-Dil	61.44	75.50	76.74	76.12	4.42	126	0.208
FastCrack-Shuffle	59.81	74.20	75.51	74.85	3.62	175	0.169
FastCrack-Ghost	61.62	73.91	78.76	76.25	3.98	122	0.1855
FastCrack-SPOS	62.88	76.07	78.38	77.21	7.08	147	0.291

Table 3. Comparison between FastCrack-SPOS and common semantic segmentation models

View table

Table 3. Comparison between FastCrack-SPOS and common semantic segmentation models

Model	IoU /%	Precision /%	Recall /%	F1 /%	FLOPs /10⁹	$S_{F P S}$ /（frame·s^-1）	Params /10⁶
DeepCrack	23.23	31.24	54.41	39.69	2190.8	1.6	30.91
HED	59.91	73.04	76.92	74.93	321.5	10.2	14.72
UNet	58.99	74.49	73.93	74.21	669.75	5.1	31.04
SegNet	62.46	77.22	76.57	76.89	1205.34	4.46	16.31
DeeplabV3+	62.56	73.41	80.89	76.97	360.25	7.7	59.34
ESPNet	61.89	75.22	77.73	76.46	88	99.2	0.364
DFANet	48.21	62.77	67.52	65.06	7.18	47	2.177
FastSCNN	61.74	76.86	75.83	76.34	3.48	132	1.136
FastCrack-SPOS	62.88	76.07	78.38	77.21	7.08	147	0.291

Table 4. Generalization experiment of FastCrack-SPOS on tunnel crack dataset

View table

Table 4. Generalization experiment of FastCrack-SPOS on tunnel crack dataset

Method	IoU /%	Precision /%	Recall /%	F1 /%
DeepCrack	12.87	20.07	32.57	21.21
HED	20.27	25.24	50.77	33.72
UNet	19.88	41.20	27.76	33.17
SegNet	20.18	28.92	40.04	33.59
DeeplabV3+	21.91	28.01	50.15	35.94
ESPNet	20.38	26.43	47.10	33.86
DFANet	9.53	11.14	39.80	17.41
FastSCNN	20.68	28.89	42.13	34.28
FastCrack-SPOS	23.64	38.52	43.21	40.73

Tools

Get Citation

Copy Citation Text

Zhuang Yue, Xiaodong Chen, Yi Wang, Huaiyu Cai, Weixi Yan, Liying Hou. FastCrack: Real-Time Pavement Crack Segmentation[J]. Laser & Optoelectronics Progress, 2023, 60(6): 0615006

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites