Caltech Pedestrian Detection Benchmark Dataset is a dataset for detecting pedestrians. It contains about 10 hours of 640*480 30Hz videos, mainly shot by cars driving on rural streets. The videos total about 250,000 frames, including 350,000 bounding boxes and annotations of 2,300 pedestrians. The annotations include the correspondence between the bounding boxes and detailed occlusion labels.

The dataset was released by Caltech in 2009 and has been modified several times since then.