Enhancing the Performance of a Rainfall Measurement System Using Artificial Neural Networks Based Object Tracking Algorithms

Abstract

With the recent development of optical sensing and digital image processing techniques, high-speed cameras have been applied to measure the microphysical properties of raindrops. However, the performance of such systems are significantly affected by object tracking algorithms. In order to improve the measurement accuracy of rainfall rate and accumulated rainfall, a novel object tracking algorithm based on artificial neural networks (ANN) is proposed in this paper. The ANN model takes the features of each raindrop in the two successive images as inputs including the center coordinates, area, canting angle, the lengths of long axis and minor axis of the equivalent ellipse. The output of the ANN model is the matched probabilities of each pair of raindrops between before and after images. Experimental data were collected during a real rainfall event. Performance comparisons between the traditional and ANN based object tracking algorithms are conducted based on the experimental data. Experimental results suggest the successful matching rate is significantly increased from 87.20% to 95.60% due to the usage of the ANN based algorithm. Hence, the improved disdrometer system is capable of producing more accurate and robust measurements of rainfall status.