Low-cost coastal mapping using Kinect v2 time-of-flight cameras

Abstract

High-resolution elevation and bathymetry data for coastal zones is extremely valuable to many researchers, however the cost of acquiring such data is prohibitively expensive for most research budgets, as it relies on specialized hardware. Mass produced off-the-shelf consumer cameras and sensors are becoming increasingly powerful, and can be affordable alternatives for collecting data. Microsoft's original Kinect sensor was repurposed to collect data for Earth sciences research, but its low depth resolution hindered its usefulness for creating accurate maps. In this paper, we evaluate Microsoft's next generation Kinect for Windows v2 sensor, which employs time-of-flight technology. Based on our results, the new sensor has great potential for use in coastal mapping and other Earth science applications where budget constraints preclude the use of traditional remote sensing data acquisition technologies.

Kinect, a Novel Cutting Edge Tool in Pavement Data Collection

“…The resolution for color camera which is 1920*1080 is also higher than the previous version of Kinect, but the accuracy of depth which is taken by Kinect V2 is found to be the same as the first version (the standard deviation of depth is less than 3.5 mm at 4.5 meters' distance) (Khoshelham & Elberink 2012). The color camera's field of view is a bit more than a depth image (84.1 degrees in Horizontal and 53.8 degrees in Vertical) (Butkiewicz 2014). The Kinect V2 has a features that can map the pixels from color image to pixels from depth image.…”

“…Khoshelham and Elberink in 2012 showed that by increasing the distance between Kinect and the surface, the random error of depth measurement will increase quadratically from a few millimeters to 4 centimeter (in the 5 meter distance) (Khoshelham & Elberink 2012). The tool is so cheaper for collecting RGB-D data rather than similar tools like Laser sensors which cost up to several thousand dollars (Yang & Diez-Roux 2012;Butkiewicz 2014). It can record up to 30 frames per second which equals to saving more than 540 million pixels per minute (30Hz frequency).…”

“…It can record up to 30 frames per second which equals to saving more than 540 million pixels per minute (30Hz frequency). The Kinect V1 measured the depth by spreading a known pattern of infrared dots at 830 nm inside the range of its field of view(57 degrees Horizontal, 43 degrees Vertical,  27 degrees physical tilt range) from infrared laser emitter into the surface and capturing(recording) them by the infrared CMOS sensor (Mankoff & Russo 2013;Butkiewicz 2014). The temperature of laser is constant during the projection by a Peltier element but it's useful to mention that thermal changes would cause an inaccuracy in the captured data of Kinect (Mankoff & Russo 2013;Butkiewicz 2014).…”

“…The Kinect V1 measured the depth by spreading a known pattern of infrared dots at 830 nm inside the range of its field of view(57 degrees Horizontal, 43 degrees Vertical,  27 degrees physical tilt range) from infrared laser emitter into the surface and capturing(recording) them by the infrared CMOS sensor (Mankoff & Russo 2013;Butkiewicz 2014). The temperature of laser is constant during the projection by a Peltier element but it's useful to mention that thermal changes would cause an inaccuracy in the captured data of Kinect (Mankoff & Russo 2013;Butkiewicz 2014). The infrared camera can't detect the IR pattern in returning phase in the sunlight because "the sun saturates the sensor" (Mankoff & Russo 2013), so the applicability of Kinect is limited to the environmental without direct sunlight.…”

“…This fast clock signal, flash the area intermittently with an array of three laser diodes which emit through diffusers with short pulse of infrared light. The distance is determined by the reflecting pulse into the camera (Demerjian 2013a;Demerjian 2013b;Lau 2013;Butkiewicz 2014 There are some mathematically based techniques for calibrating and matching the RGB-D data which can be found in the literature. (Xu et al 2011;Tong et al 2012;Herrera et al 2012;Smisek et al 2013).…”