Abstract: Recent advances in computer vision and deep learning are allowing researchers to develop automated environment recognition systems for robotic leg prostheses and exoskeletons. However, small-scale and private training datasets have impeded the widespread development and dissemination of image classification algorithms (e.g., convolutional neural networks) for recognizing the human walking environment.