Vision Transformer Customized for Environment Detection and Collision Prediction to Assist the Visually Impaired

. 2023 Aug 15;9(8):161. doi: 10.3390/jimaging9080161

Algorithm 1: Collision avoidance and feedback

1.
Capture video from the camera
2.
$for i = 0, 1, \dots, N, N = NUMBER OF BOUNDING BOXES IN ONE FRAME do$
3.
Store the center point, height, and width of the bounding box in the current frame, and the object trajectory vectors
1. $[x_{I_{c (t_{i})}}, y_{I_{c (t_{i})}}] = [\frac{x_{I_{(t_{i})}} + w_{I_{(t_{i})}}}{2}, \frac{y_{I_{(t_{i})}} + h_{I_{(t_{i})}}}{2}]$
2. $a = x_{I_{c (t_{i + 1})}} - x_{I_{c (t_{i})}}$
3. $b = y_{I_{c (t_{i + 1})}} - y_{I_{c (t_{i})}}$
4.
Check if the object is moving toward the user:
- a.
  $If w_{I_{(t + 1_{i})}} \times h_{I_{(t + 1_{i})}} > w_{I_{(t_{i})}} \times h_{I_{(t_{i})}} then$
  1. $If x_{I_{c (t_{i})}} < \frac{w_{i m g}}{3} A N D b < 0$
    
    print “Name of the detected object” is approaching you from your left side!
  2. $elseIf \frac{w_{i m g}}{3} < x_{I_{c (t_{i})}} < \frac{2 w_{i m g}}{3} A N D b < 0$
    
    print “Name of the detected object” is in front of you!
  3. $elseIf \frac{2 w_{i m g}}{3} < x_{I_{c (t_{i})}} A N D a < 0 A N D b < 0$
    
    print “Name of the detected object” is approaching you from your right side!
- b.
  $If w_{I_{(t + 1_{i})}} \times h_{I_{(t + 1_{i})}} = w_{I_{(t_{i})}} \times h_{I_{(t_{i})}} then$
  
  print “Object is static!”
- c.
  $end$
5.
Store the current center points and distance for the next iteration

$[x_{I_{c (t_{i})}}, y_{I_{c (t_{i})}}] = [x_{I_{c (t + 1_{i})}}, y_{I_{c (t + 1_{i})}}]$
6.
$end$