This paper proposes an Ambient-Assisted Living (AAL) system that harnesses the power of Deep Learning to provide enhanced support for elderly disabled individuals in smart homes. The system introduces an Improved Disp R-CNN based approach to activity recognition and fall detection, which offers several advantages. It can identify specific actions such as meal preparation or medication administration, while also accurately detecting human body parts and their movements through displacement analysis. The system relies on strategically placed CCTV cameras to capture video data, which is then analyzed in real-time by the trained Improved Fall Detection model. To ensure ongoing accuracy, the model can adapt to the unique needs and movements of each resident. By identifying potential safety hazards like falls and monitoring deviations from established activity patterns, the system actively promotes the well-being of elderly residents. Real-time alerts and interventions are triggered based on these detections, further enhancing their safety and security. Moreover, caregivers and family members can remotely monitor residents through an intuitive interface, enabling them to provide essential support. The proposed method achieved an accuracy of 0.92, demonstrating its effectiveness in fall detection. This advanced Improved Fall Detection model based AAL system represents a significant breakthrough, as it facilitates personalized care and significantly improves the quality of life for elderly residents with disabilities, effectively addressing the challenges posed by an aging population.