Visual Identification of Mobile App GUI Elements for Automated Robotic Testing

Abstract

Automated robotic testing is an emerging testing approach for mobile apps that can afford complete black-box testing. Compared with other automated testing approaches, automatic robotic testing can reduce the dependence on the internal information of apps. However, capturing GUI element information accurately and effectively from a black-box perspective is a critical issue in robotic testing. This study introduces object detection technology to achieve the visual identification of mobile app GUI elements. First, we consider the requirements of test implementation, the feasibility of visual identification, and the external image features of GUI comprehensively to complete the reasonable classification of GUI elements. Subsequently, we constructed and optimized an object detection dataset for the mobile app GUI. Finally, we implement the identification of GUI elements based on the YOLOv3 model and evaluate the effectiveness of the results. This work can serve as the basis for vision-driven robotic testing for mobile apps and presents a universal approach that is not restricted by platforms to identify mobile app GUI elements.