An Autonomous Drone for Measuring A River Width Based on K-Means Clustering

The common method for measuring the river width is by getting two GPS coordinates from both sides of the river and calculating the distance in between. However, it is inefficient and time consuming. This paper proposed a new method by using a drone with special performance to measure the river width based on K-Means Clustering. The drone is equipped with a camera and minicomputer of Jetson Nano, with the use of Pixhawk to control the motors and stabilize drone poses. The Jetson Nano is connected to the Pixhawk via a USB cable; hence it makes Jetson Nano easily accessing the Pixhawk data such as the IMU, the GPS coordinates, and the altitude. The river width is calculated based on the color similarity of the color of water, it is extracted by K-Means clustering, then the actual river width is obtained by normalization of the river width on pixel image and drone altitude. The proposed system has been tested in real experiments with a drone flying over rivers which have five different conditions. The first test is applied to a river with sufficient water conditions, the river edge has a sloping concrete of ±7.2m wide, and an optimal height of 20m. The test has a reading error of 3.1944%. The second test is applied to a river with a sediment of±5m, and an optimal height of 35m, it has a reading error of 2%. The third test is given to a river with slightly winding and sufficient water, the optimal altitude is 35m and it has a reading error of 1%. The fourth test is given to a river with a lack of water, uneven edges, sediments, and a width of ±9m. The optimal altitude is 45m, and it gives a reading error of 0,67%. The fifth test is a river with an irregular bank of land and width of ±7.7m. The optimal altitude is 40m, and it gives a reading error of 1,039%.