A comparative analysis of four object detection models YOLOv11, DETR, CenterNet, and Faster R-CNN was conducted for real-time assistive technology designed to support individuals who are blind or have low vision. Despite advances in computer vision, many assistive tools still struggle with real-time performance, particularly in dynamic, cluttered environments. Most existing studies focus on general use cases, often neglecting accessibility-specific needs. This study fills that gap by evaluating one-stage, transformer-based, anchor-free, and two-stage models using a real-world dataset. YOLOv11 achieved the best balance of speed and accuracy, with a mean Average Precision (mAP) of 0.52 and an inference time of 2.6ms, making it ideal for edge deployment. Faster R-CNN delivered the highest precision but suffered from slower inference, limiting its usability in real-time scenarios. These results underscore the importance of tailoring detection models for assistive use, balancing precision and speed to enhance accessibility solutions for the visually impaired.