Accessibility barriers remain widespread in mobile applications, limiting equitable access for over one billion people with disabilities, older adults, and users with temporary or situational impairments. While the Web Content Accessibility Guidelines (WCAG 2.1/2.2) define clear principles, many Android apps still violate essential criteria due to limited tool support and the difficulty of automated detection. This paper presents an approach along with a tool, named A11yArgus, that integrates DroidBot for dynamic exploration and UI Automator for interface extraction, combining geometric reasoning and Optical Character Recognition (OCR) to detect accessibility issues. A11yArgus identifies 13 accessibility issues, including six novel ones, covering visual, structural, and semantic aspects. We evaluate our approach and tool on 200 apps from \textit{F-Droid} and \textit{APKPure}. In particular, we achieved 91.9% precision, 97.8% recall, and an F1-score of 94.8%. Our results reveal persistent violations at the WCAG Level~AA and emphasize the importance of visually grounded and context-aware analysis for advancing inclusive mobile software.