The detection of latent fingerprints plays a crucial role in criminal investigations and biometrics. However, conventional techniques are limited by their lack of depth-resolved imaging, extensive area coverage, and autonomous fingerprint detection capabilities. This study introduces an object-driven optical coherence tomography (OD-OCT) to achieve rapid, autonomous and ultra-large-area detection of latent fingerprints. First, by utilizing sparse sampling with the robotic arm along the slow axis, we continuously acquire B-scans across large, variably shaped areas (∼400 cm2), achieving a scanning speed up to 100 times faster. In parallel, a deep learning model autonomously processes the real-time stream of B-scans, detecting fingerprints and their locations. The system then performs high-resolution three-dimensional imaging of these detected areas, exclusively visualizing the latent fingerprints. This approach significantly enhances the imaging efficiency while balancing the traditional OCT system's trade-offs between scanning range, speed, and lateral resolution, thus offering a breakthrough in rapid, large-area object detection.