Real-Time Computing for Droplet Detection and Recognition

Abstract

In this paper, we propose a novel framework to achieve real-time droplet detection and recognition. In a fluidics system, water drops can be generated to carry objects. It is important to use such droplets to analyze the properties of the carried tiny objects. In real applications, the movement of droplets are captured by a high-speed camera, and processed by an image-based computing system. This imposes high requirements on the computational efficiency of the image processing and pattern recognition methods in the system. To tackle this challenge, we propose to use efficient object tracking and classification methods to build the framework. We first use the Hough transform as the basic droplet detection method, and then explore the relationship of the same droplet in different frames to track their movement. For the classification step, we use the area of particle content as the basic feature, and conduct classification using a linear support vector machine. The experiments on real-time droplet analysis system verifies that our proposed framework can process images at 1000 frames per second, while achieving 99.3% in classification accuracy.