Final Year Project 2 - Progress 2

1. The pre-filmed video is fed into the model as the input. The first frame of the input video is captured as shown in Figure 1.
2. Four points are selected from the first frame image. These points are connected with red lines forming a closed area, which is the desired region for person detection. 
3. Figure 3 illustrates the video frame that has been transformed into 2D bird's view. 
4. The bird’s view video is then fed into the YOLOv4 pre-trained model frame by frame for person detection. Figure 4 shows the output of the detection model.
5. The bounding boxes represented in green indicate the person is maintaining a physical distance as the distance is within the acceptable threshold value. The people with red bounding boxes indicate that they violate the pre-defined threshold value. On the upper right of the frame, the number of people violating the physical distance is shown.

Figure 1: First frame of the input video

Figure 2: Frame with selected four points connecting with line

Figure 3: 2D bird’s view generated from the Perspective image

Figure 4: Output of physical distancing detection model using bird’s view video frame