Lane keeping is a fundamental function for self-driving vehicles, ensuring they can follow road lanes accurately and safely without human intervention. While many existing approaches rely either on computationally expensive CNN-based end-to-end models or shape-supervised algorithms that require specialized sensors, both are difficult to deploy on embedded or small-scale platforms. This paper presents the design and implementation of a vision-based Lane Keeping Assistance (LKA) system deployed on a 1:10 scale self-driving vehicle. Using Stanley algorithms, the LKA system continuously detects lane boundaries and applies steering corrections to maintain safe positioning. A Traffic Sign Recognition module is also integrated to simulate real-world driving and enhance contextual decision-making using a lightweight object detection network, allowing the vehicle to respond appropriately to signs. Unlike prior works that evaluate perception and control separately, our novelty lies in the unified integration of lane detection, control, and sign recognition into a real-time pipeline optimized for resource-constrained hardware. The integration of these systems is evaluated under various environmental and road conditions to assess their accuracy and robustness.