YOLOFOR: YOLO AND OPTICAL FLOW FOR FORENSIC ANALYSIS OF ATTACKED DRONE CASE

Drone devices have been used for various activities recently. During its op-eration, a drone may get into an accident, and the authority needs to find the root causes. For this purpose, a forensic analysis is needed to collect relevant information to help and support the investigation. Several evidentiary artifacts can be utilized, such as images or videos captured from the drone’s camera. Furthermore, from the video data, it can be seen whether the object causing the incident is moving toward the drone, or the drone is actually moving toward the object by estimating the direction of the object’s move-ment. In this study, we propose YOLO and Optical Flow for Forensics (YOLOFOR), an attack detection method that can perform object detection accompanied by an object’s movement direction estimation model to assist forensic investigation on drone video data. The proposed method comprises two main components, specifically YOLOv5 and Lucas Kanade. YOLOv5 is used for object detection, while Lucas Kanade is utilized to calculate the direction of moving objects. The experimental results demonstrated that the proposed method could detect objects with an mAP of 0.633. The estimated direction produced by Lucas Kanade has the least number of the quiver, indicating that the quiver is more fo-cused on the detected moving object.