American Sign Language (ASL) enables vital communication for deaf and hard-of-hearing individuals, yet automated recognition of its gestures remains challenging. Here, we assess three leading object detection frameworks, YOLOv11, Faster R-CNN and RT-DETR, on a moderately sized, extensively augmented ASL dataset. By comparing average precision across a range of overlap thresholds (mAP 50–95), as well as measuring inference latency and computational cost, we show that YOLOv11-l strikes the best compromise between speed and accuracy for real-time use. In contrast, RT-DETR-x attains the highest overall precision but demands substantially greater resources. Our results clarify how each model’s trade-offs affect performance and lay the groundwork for refining ASL gesture detection. This work brings us closer to practical, responsive systems that can seamlessly interpret sign language in everyday settings.