In order to supplement the lack of training data for object detection purpose, synthetic data was generated by using Unity Perception package with strategies of domain randomization. YOLOv7 was used as the object detection model and up to 37% improvement was achieved.