Fall detection in automated buses using in-vehicle camera -basic study using support vector machine-

To enhance in-vehicle safety in automated buses, a system that can detect whether falls have occurred in the bus can support the remote human controllers in quickly responding to the running bus. In this study, we designed a fall detection system that utilizes Wong Kin Yiu’s yolov7 for extracting features from the video data. Then support vector machine was used for the discriminant task of whether a fall has occurred or not given an image as the input data. We tested our system with various settings, including full features versus reduced features, and raw images from the camera versus images that contain full views of the human inside the bus, all with multiple camera views. And finally, we confirmed that the 360-degree camera could detect human falls. Using several evaluation criteria, such as accuracy, precision, recall, and F1 scores, we showed how the detection ability of the system varied according to different settings of the test. The highest F1 score of 0.980 was achieved in the experiment. In addition, we found that if a full-body view of a human being could be achieved, the accuracy of the fall detections would be greatly improved. Finally, we discuss several potential directions for further research of this topic.