As the number of personal mobility users increases, it is essential to protect both personal mobility users and car drivers. Although existing supervised learning-based object detection methods can recognize personal mobilities, processing for data collection, labeling, and training is time-consuming and expensive whenever new forms of personal mobilities come to the market. Anomaly detection-based methods are proposed to learn normal patterns and detect abnormal patterns without prior training. This study proposes a new approach to personal mobility user recognition using deep learning-based anomaly detection in dynamically changing driving environments. The proposed approach consists of Human detection, Cropping and Anomaly detection modules. The Human detection module detects human regions. The Cropping module removes unnecessary areas and performs a preprocessing. The memory-based Anomaly detection module distinguishes between pedestrians and personal mobility users. Based on the proposed anomaly detection method, augmented reality (AR) visualization in the head-up display (HUD) is proposed for effective driving assistance. Since it is possible to distinguish pedestrians and mobility users effectively, the AR HUD-based visualization can assist the driver to pay more atten...