A reflection-based dataset for object and human detection

In addition to being a critical task in and of itself, detecting an object’s reflection on another reflecting surface makes the work of detecting objects or people a little more difficult because cameras can avoid reflections due to noise, specularity, focusing, and shining. In this work, a specialized dataset of photographs of people and other objects reflecting off of various reflective surfaces—such as mirrors, glass windows, steel surfaces, and eyeglass lenses—is presented. High-resolution cameras were used to gather the dataset mostly on a college campus at different times of day and in a variety of lighting situations. To mimic the diversity of real-world environments, it incorporates pictures of people and objects in reflection that were taken from various perspectives.