Image recognition technology has been widely used in urban public monitoring systems. This study aims to utilize monitoring systems to locate visually impaired individuals in public spaces. The authors established a dataset of visually impaired individuals across multiple scenes and employed mainstream object recognition systems for training and testing purposes. Through experiments, the majority of models demonstrated the capability to accurately recognize visually impaired individuals in arbitrary scenes. For instance, commonly used systems such as YOLOv4 and YOLOv5 can achieve detection accuracy of over 92% (mAP50). The experimental results show that the dataset provided in this research can be used for effectively recognizing visually impaired people’s features when applied to mainstream object detection systems in multiple scenes.