Currently, AI edge device is becoming the trend in real-time object detection. However, one of the disadvantages of current object detection models is the limitation of objects or the lack of additional data for existing objects. Therefore, labeling data is an important task. This study proposes a new method for labeling object detection in video, which collects from moving cameras using edge devices. Specifically, our method is able to collect sharp resolution frames containing new objects and objects that are mis-detected during the real-time running of AI Edge devices. The application of this solution supports locating new objects and suggests adding data to existing data in a frame/image, which is able to save a lot of time and effort for labeling video data.