查看文章

Double pendulum exhibits a rich dynamic behavior and is often used to study nonlinear dynamics, instability, and chaos. In order to devise a method for its control, the angles and the angular velocities of its two arms need to be measured accurately with sufficiently fast sampling rate. Generally, encoders and other sensors are used which are costly, difficult to install for custom built pendulums, and require continuous calibration to measure the angles. Therefore, in this paper we present a comparison of four image processing based methods to estimate the angles of double pendulum. The first two methods are widely used involving variants of Canny edge detection in conjunction with Hough transform. The third method extracts the skeleton of the thresholded pendulum and then applies Hough transform on the skeleton to compute the angles. The fourth method, never before used for double pendulum, segments the pendulum arms and uses basic matrix operations to extract the angles. The results focus on the speed-accuracy-memory tradeoff for all the methods and manifest that the fourth method is satisfactorily accurate with minimum computational complexity and memory usage.