When Paul and I first began work at i-TEN Associates in 2009, we perceived photogrammetry as a soft science, a dying industry, and felt that its role could easily be replaced by more traditional methods of survey. I couldn’t begin to imagine how two photographs put together in a stereo pair could produce xyz data, let alone data that was accurate enough to be used for projects that called for a high-precision end product. I didn’t understand why anyone would choose to use photogrammetric services over traditional survey methods. Flash forward to today and my opinions are now quite the opposite.
My eyes were opened to the value of this science while working as a sub-contractor on a project for the Army Corps of Engineers. ACOE had hired a local firm to survey a sediment retention structure (SRS) but time and budget made it impossible for the firm to survey and map the entire project area. i-TEN was hired to provide the broader base map because of our ability to create a surface model in a short period of time. The survey firm would survey the immediate site of the SRS because of the need for a surface with higher accuracy.
While the survey firm performed their preliminary round of site shots, i-TEN had the site flown to acquire our aerial imagery. Back at the office, we created an Aerial Triangulation (AT) solution based upon ground control points already established by the surveyor and began the process of collecting break lines and mass points through stereo compilation. From the stereo imagery i-TEN has the ability to visualize survey point locations in 3D space – which historically has revealed errors in the survey documentation process such as ‘floating’ control due to transposed numbers. Eliminating erroneous survey control data in this way provides a check and balance system for traditional survey and can prevent on-site financial disasters such as unnecessary excavation and damage to delicate ecosystems.
As the project developed, comparisons between our surfaces generated through photogrammetric collection and the surfaces generated through total station survey revealed the Triangulated Irregular Network (TIN) created from survey data had very large spanning triangles which were the result of the distance between survey points. Our triangles were small and tightly clustered. As a result of this dense data, we were able to compile our solution in a comparable amount of time as the traditional survey which brought up the subject of ‘relative’ accuracy.
Traditional survey methods can typically come within 1.5mm +/- 2 ppm (Wiki). From the 2.5 cm ground sampling distance digital imagery acquisition, we were able to obtain 0.10 ft. (3.048 cm) vertical accuracy on hard surfaces for the base mapping topography data. Clearly, 1.5 mm is more accurate than 30 mm. However, within the context and scale of the project, though the surveyed individual points themselves were more accurate, the surface data that was interpolated between the points becomes less and less accurate with the diminished density of the data. In this image, pink represents land survey data, while cyan represents photogrammetric ground sampling. Note the triangulated difference in density.
An important question arose from this project: ‘Is a surface that is generated from a few very precise points really more ‘accurate’ than a surface generated from many slightly less accurate, but much denser points?’ In this context, the surface generated from survey data was in fact less accurate than the surface generated from photogrammetric collection.
In another, somewhat controversial series of projects within Oregon University Systems, university recreation centers are going through a sweep to be remodeled – funded by increased tuition of students (that’s the controversial part). i-TEN was contracted to laser scan and 3D model an area of interest and ensure that scans were oriented to local control points established by an area surveying firm.
From the established survey control, partial 3D modeling had been accomplished using Autodesk Civil 3D for areas of interest such as curbs and partial exterior walls. Once laser scanning was complete and set to control, we discovered 3D model spatial offsets of up to 6" between the pre-modeled data interpolated from survey, and the superimposed point cloud. The cause for this was not surveyor error but rather a limitation on the number of points available and inferences. Therefore, we can say with confidence from project experience that the relative representation of surface accuracy in this case as well, is improved with higher density of surface control data.
To the question of "accuracy": Is a surface solution from many slightly less precision points more accurate than one created from a few high-precision points? In these scenarios, our experience tells us yes.