GENERATION OF HIGH RESOLUTION DSM USING UAV IMAGES
Introduction
Conclusion
Post Processing of UAV images are better supported by algorithms from Computer
Vision: SIFT Algorithm for feature extraction ,Dense Stereo Matching for Image
Matching etc.
Among the range of available commercial software packages , PIX4D provides
more options for optimizing the results.
For areas with smaller spatial coverage , UAVs images provides best data to
generate high resolution photogrammetric products.
Project Member: Uttam Pudasaini (011809-10) , Biplov Bhandari (011793-10), Niroj Panta (011807-10), Upendra Oli (011806-10)
Project Supervisor: Mr. Uma Shankar Panday Project Co-supervisor: Asst. Prof. Nawaraj Shrestha LMTCKU
Recommendations
High end work station is required for processing large number of UAV images
through new algorithms.
Make use of well distributed and accurately measured check points for assessing
the accuracy of DSM.
Object oriented image analysis techniques can be explored for increasing the
accuracy of final DSM
LPS-PIX4D LPS-AgiSoft PIX4D-Agisoft
Classical Photogrammetric Workflow Computer Vision Workflow
Particulars
Aerial
Photogrammetry UAV Photogrammetry
Data Acquisition Manual/Assisted Assisted/Manual/
Automatic
Aerial Vehicle Highly stable
specially designed
aircrafts
Small aerial Vehicles
with certain payload
capacity
GPS/INS
Configurations
cm-dm level
accuracy
cm-10 m
Image Resolution cm-m mm-m
Ground Coverage Km2 m2-km2
Cameras Well calibrated
cameras especially
designed for
photogrammetric
applications
Can work with normal
digital cameras
Fudicial Marks Present Absent
Flying Height 100 m-10 km m-km
(not more than 1 km)
Data Processing
Workflows
Standard
Photogrammetric
Workflow
No standard workflows
Salient Feature Better control over
the output image
quality
High temporal accuracy
with real time
applications
Digital Surface Models (DSM):
Digital representation of the
earth’s surface elevation
including natural and artificial
objects like trees or building
above it.
UAV Photogrammetry:
Provides a low cost
photogrammetric platform.
An emerging field that can
provide very high resolution
datasets for small areas.
Remotely or (semi)
autonomously controlled
without human pilot.
Among the range of terrestrial and aerial methods available to produce high resolution datasets, this project tests the utility of images acquired by a fixed wing, low cost Unmanned
Aerial Vehicle (UAV) by making use of image processing algorithms ranging from classical photogrammetry to modern Computer Vision (CV) algorithms.
The effort and the achievable accuracy of DSM resulted from every process are compared using the highly accurate ground control points as the reference data. The comparison of the
DSM is performed through difference of DSM, RMSE and visual interpretation. Although three software: LPS, AgiSoft PhotoScan and PIX4D were used for image processing, the
identified algorithms and limitations in processing are valid for most other commercial photogrammetric software available on the market.
Objectives
Main Objective
To create a high resolution DSM using images acquired by a digital camera
mounted in a UAV platform.
Software Used
Sub-Objectives
To orient and georeference UAV images using internal and external orientation
parameters.
To generate Digital Surface Model.
To compare and analyze the accuracy of DSM generated from different methods
Data Used
27 high resolution images acquired by a Trimble
UX5 Imaging Rover
(2.4 cm average spatial resolution)
Control Points
(GCP +Check Points)
Camera
Calibration
parameters:
Focal
Length,
Pixel Size
and
Distortion
Parameters
Abstract
Larger spikes on output DSM from LPS.
Classical Photogrammetric image
matching algorithms fails for areas with
homogenous and repetitive pattern.
Poor results at area covered with trees
and vegetation.
For mixed topography, all the algorithms
works fine.
PIX4D provided the best result in all
cases.
.
Visual Interpretation and Analysis
Point
No
Elevation (m) Elevation Difference (cm)
Original
(O)
DSM
LPS
(a)
DSM
AP
(b)
DSM
PIX4D
(c) O-a
O-b
O-c
2003 136.173 136.278 136.116 136.201 -10.54 5.66 -2.84
2004 128.362 128.392 128.422 128.375 -3.04 -6.04 -1.34
2006 132.402 132.262 132.381 132.382 13.960 2.06 1.96
2007 127.585 127.649 127.653 127.571 -6.44 -6.84 1.36
2010 131.953 132.052 131.941 131.962 -9.94- 1.16 -0.94
RMSE(cm) 9.546 4.917 1.813 Mean=
8.783
Mean=
4.348
Mean=
1.688
RMSE Computation Difference of DSM
DSM generated from LPS DSM generated from PIX4DDSM generated from AgiSoft
DSM Generation
Aerial Imagery Camera Parameters
Image Matching
Georeferencing
Interior Orientation
Exterior Orientation
Aerial Triangulation
Bundle Block adjustment
Interpolation
DSM Generation
UAV-acquired Imagery a. Orientation Parameters
b. GCPs
1. Initial
Processing
Image Matching
(Between Images)
Automatic Aerial
Triangulation
Bundle Block
adjustment
Image by image
Key point
Extraction
Densified Point
Cloud
2. Point Cloud
Densification
Filtered Point
Cloud
3. DSM
Generation
Aerial Photogrametry VS UAV Photogrametry
