Robust object tracking in unconstrained environments remains a critical challenge due to occlusion, motion blur, background clutter, and target deformation. This paper presents LTMOT, a Long-Term Memory-Aware Object Tracking framework that leverages recurrent memory and attention mechanisms for enhanced tracking accuracy and temporal stability. Our model integrates a lightweight CNN backbone for spatial feature extraction, an LSTM-based memory unit to retain long-term contextual information, and a cross-frame attention module for adaptive focus on target-relevant regions. We conduct extensive experiments on multiple benchmarks, including OTB-100, LaSOT, and TrackingNet, where LTMOT consistently outperforms state-of-the-art trackers in terms of precision, success rate, and real-time capability (achieving 32 FPS). Ablation studies further confirm the complementary benefits of memory and attention in maintaining performance during prolonged occlusion and changes in appearance. The proposed tracker demonstrates strong generalization and efficiency, making it suitable for real-world deployment in various applications, including surveillance, robotics, and autonomous systems.