Forensic Analysis of Car Accident Using MobileNet and Optical Flow

Accident investigation is an effective aspect of reconstructing car accident cases by utilizing video recorded by a third party using closed-circuit television (CCTV). The video is processed using computer vision and analyzed to determine the perpetrator or object whose movement is abnormal or anomalous. One by one, frames from CCTV are examined to identify anomalies in the movement of car objects on the highway that were caught on CCTV cameras. This is performed to ascertain the cause of the car accident. In this paper, we propose the use of the MobileNet model to classify car accident images extracted from CCTV. In addition, we apply the optical flow technique to detect flow for object direction movement. This paper combines MobileNet and optical flow methods to obtain high performance in car accident classification. The MobileNetV2 model in the collision class achieves precision of 97.27%, recall of 98.94%, and F1 score of 98.10%. In addition, the optical flow model shows that the MAPE value is 1.82%.