Algorithm 2. Copy-Paste Objects from Multiple Images

Input:    - ‘results’: Output of Algorithm 1    - ‘fallen_person_imgs’: List of fallen person images Output:    - Augmented image

For each ‘bg_img’ in ‘bg_imgs’ do    Initialize an ‘occlusion_mask’ to 0    Iw, Ih = size of ‘bg_img’    Open corresponding file in ‘bg_txts’, read lines into ‘bg_lines’    For each ‘bg_line’ in ‘bg_lines’ do       Update the corresponding ‘occlusion_mask’ to 255    If the section changes, then       Update ‘heightmax’, ‘widthmax’, ‘heightmin’, ‘widthmin’, ‘areamax’       Create a ‘bonnet’ mask with Grounded-SAM    ${RoI}_{ratio}=\left|\frac{{(I}_w-{bbox}_{min\_x})- {bbox}_{min\_x}}{{(I}_w-{bbox}_{max\_x})- {bbox}_{max\_x}}\right|$    If ‘${RoI}_{ratio}$’ > 0.5, then       ${RoI}_{ratio}=1-{RoI}_{ratio}$       ${RoI}_{thr}=\left|{bonnet}_y-b{box}_{min\_y}\right| * {RoI}_{ratio}+{bbox}_{min\_y}$    For each ‘fallen_person_img’ in ‘fallen_person_imgs’ do       Break if you encounter the maximum paste object value during the loop       Create a ‘fallen person’ mask with the YOLOv8x-seg       Adjust position and size for fallen person within RoI       Randomly rotate ‘fallen person’ with a probability of 0.25       Each ‘fallen person’ is flipped vertically and horizontally with a probability of 0.5       if ‘occlusion_mask’ exists at the current position, then          continue       Paste a ‘fallen person’ into the background image       Apply alpha blending    Save the augmented image