<p>Simultaneous Localization and Mapping (SLAM) in dynamic environments presents a significant challenge, especially for resource-constrained systems. This challenge is amplified when using low-resolution cameras, which impede robust feature extraction and tracking. While many dynamic SLAM systems achieve high algorithmic precision, their performance on low-power hardware with low-quality sensors is often not a primary focus. To address this gap, we propose LR-SLAM, an efficient vision-based SLAM system designed for robust operation in dynamic indoor environments using low-resolution RGB-D cameras. To mitigate poor image quality, LR-SLAM employs a GCNv2 network for enhanced feature extraction and an adaptive non-maximum suppression algorithm to promote a uniform distribution of feature points, thereby improving tracking stability. The system also integrates a multi-stage dynamic feature rejection strategy, combining lightweight object detection with epipolar constraints and a probabilistic model to effectively identify and remove features on moving objects. Extensive testing on the public TUM dataset and in real-world scenarios demonstrates that LR-SLAM achieves high localization accuracy and robustness, operating in real-time at an average of **47 ms per frame**. By operating effectively with low-resolution cameras, our work presents a practical solution for dynamic SLAM and broadens its application potential on resource-constrained platforms.</p>