| Nowadays, mobile phone has become more and more common, and it seems to be a necessity of life with the endless emergence of various mobile applications. In order to ensure the quality of the user experience, software testing is a very important procedure. However, traditional automated testing methods are not able to meet the growing needs of the current test due to the improvement of phone features, applications complexity and the rising labor cost. To solve some of the problems unhandled by traditional automatic testing, this thesis utilized one camera to shoot the operational process of a phone, then handled the pictures captured with image processing and machine learning algorithms, and sent corresponding instructions to the phone afterwards. This thesis mainly focuses on the automated testing algorithms of game applications, enabling these applications to play automatically without human intervention. Many game applications require manual real-time control of character moving up, down, left, and right. Furthermore, there are many scenes with complex background in game applications, which make them much more difficult to test by traditional approach.The main content and innovations of this thesis are as follows:â‘ Proposed automated testing algorithms based on handcrafted features. Combined handcrafted features with Support Vector Machines(SVMs) to get the upcoming action of the phone, which can make the computer play the APP automatically. And improved the histograms of oriented gradients to avoid early action. These algorithms can be applied to automated test of mobile applications such as Parkour and simple racing Apps.â‘¡Proposed automated testing algorithms based on deep learning. Compared with handcrafted features, the features trained by deep networks are more adaptable and save manpower. Moreover, in order to cope with time series problems, the deep network designed took full consideration of both conditional Restricted Boltzmann machine(cRBM) and RBM. These algorithms can be applied to automated test of mobile applications such as Parkour and racing Apps. |
