Camera Calibration and Homogenous Surfaces

I am a survey technician who utilizes a phantom 4 pro to perform mapping of material stock pile locations.

I have recently been witnessing datasets that are coming back with a large change between initial and optimized camera parameters 5-6%. These flights do take place in areas with vertical relief as they are at stockpile locations, but the materials are homogenous (Sand, Pellets, etc). I am also having issues getting the ATPs necessary to provide full accurate coverage from these flights.

I have seen documentation online that states to calibrate your camera once a year (to come up with a new camera set of parameters), and that you should fly flights at 100 and 175 feet, each in nadir and at 30 degree oblique and run a flight in square pattern around perimeter of a high detailed building with vertical features. The articles I have come across do not state what to do with this data once captured.

Question 1: Should the camera calibration flights be run through pix4D (All at once) and save a camera model of the optimized camera upon completion of step one? Does PIX4D have a recommended process/workflow for this?

Question 2: I note that the Initial processing recommends processing Custom:1/2 Image Scale for homogenous surfaces. However, it does not state how much accuracy is lost by doing so? Does PIX4D have estimates of this?

Question 3: If I am having issues getting enough ATPs from homogenous surfaces, should I be adjusting my flight height higher or lower (or) increasing overlap? I am performing flights at 200-300 feet currently.

Hi Joe,
If you are seeing a camera optimization error greater than 5% then you should try enabling All Prior in the step 1 calibration settings. Mapper will place more weight on the initial camera internals.

If you think the camera on your drone is fault or has been damaged then you might want to consider calibrating the camera and adding it to the database. However, I would first try All Prior before venturing down this path. If you are interested then you can read this article:

If you are processing image with homogeneous surfaces then I would keep the image scale at full. Ideally, I would fly these areas with heavy overlap. Perhaps 80-90%, not to exceed 90. Mapper needs as much image content to generate a sufficient amount of keypoints.