Fully automatic DOM generation method based on optical flow field dense image matching

Figures & data

Figure 1. Flowchart of digital orthophoto map (DOM) generation.

Figure 2. Flowchart of 3D point cloud completion.

Figure 3. Knot insertion of a B-spline curve.

Figure 4. 3D point clouds completion.

Figure 5. Flowchart of the seamline network optimization.

Figure 6. Fully automatic extraction of the seamline network in a dense area of buildings.

Table 1. Parameters for the two sets of experimental images.

Item	Beijing	Preserver Island	Vahingen
Aerial craft	Low-altitude Unmanned Aerial Vehicles (UAV)	Chinese transport 12 aircraft	Aircraft
Camera	Phase One IXU-1000	Phase One RS-1900	Intergraph/ZI DMC
Date of aerial photography	May 2016	November 2017	August 2008
Principal distance (mm)	51.212,93	89.470,90	120.000,00
Format (pixels)	11,608 × 8708	11,540 × 16,470	7680 × 13,824
Pixel size (µm)	4.6	4.6	12.0
Ground Sample Distance (GSD) (cm)	7	5	8
Relative flight height (m)	779	960	900
Longitudinal overlap (%)	60	60	60
Lateral overlap (%)	30	40	60
Number of mapping strips	4	8	3
Number of control strips	4	2	1
Number of Images	88	268	20
Number of Ground Control Points (GCPs)	20	25	0
Number of pass points	161,085	113,670	7151
Block area (km^2)	2.8 × 2.8	5.2 × 5.2	2.1 × 1.8
Maximum topographic relief (m)	60	20	170
Average ground elevation (m)	507	49	285
Image matching accuracy (μm/pixels)	±0.8 / ±0.17	±1.0 / ±0.22	±2.88 / ±0.24
Pass point planimetric/height accuracy (m)	±0.070 / ±0.065	±0.066 / ±0.073	±0.089 / ±0.103

Figure 7. Doms of ten typical surface features generated by the fully automatic method proposed in this paper.

Figure 8. Mosaic accuracy curves and histograms between differential rectification images.

Table 2. DOM accuracy for the two experimental areas.

Experimental area	DOM cell size (m)	Seam accuracy			DOM accuracy
		Number of seam lines	Maximum edge error (m/pixels)	${\bar{\mu }}_M$ (m/pixels)	Check points	Maximum point error (m/pixels)	${\bar{\mu }}_{DOM}$ (m/pixels)
Beijing	0.070	89	0.157 / 2.24	0.090 / 1.28	15	0.175 / 2.50	0.078 / 1.11
Preserver Island	0.050	511	0.119 / 2.38	0.069 / 1.39	25	0.116 / 2.32	0.057 / 1.14

Figure 9. Visual comparison of the automatically generated DOMs on the Vahingen dataset.

Table 3. Quantitative comparison of the three methods.

Method	DOM cell size (m)	Processing time			DOM accuracy
		Dense image matching (min)	DEM generation (min)	Ortho rectification (min)	Check points	Maximum point error (m/pixels)	${\bar{\mu }}_{DOM}$ (m/pixels)
Proposed	0.080	27.4	5.3	5.2	10	0.208 / 2.51	0.121 / 1.51
PhotoScan	0.080	25.1	12.1	7.4	10	0.210 / 2.62	0.139 / 1.74
SURE	0.080	31.2	4.5	6.2	10	0.190 / 2.37	0.115 / 1.44

Data availability statement

Some data that support the results and analyses presented in this paper are freely available online (URLs: https://pan.baidu.com/s/1mQCQNjW9eGaesfxsYQqcaw Password: hsly). Others cannot be shared at this time, as the data also form part of an ongoing study.