Safely Back to Work
Glass or polished surfaces show the reflection of the actual object present in the scene which may be indistinguishable by the computer software . While calculating the depth of the object in an image with reflective surface , the model may estimate wrong depth which can lead to miscalculation in social distancing .
Fig . 12 : Reflection of image in The Bean at Chicago
The deep learning model running beneath the computer software is trained to detect and classify humans in a given image . The training images consist of 2D image representations of the 3D world ; hence it is hard for such models to differentiate between a mannequin and actual humans . In some cases , it may happen that a person is standing next to a mannequin and the system misclassifies it as a social distance violation .
Fig . 13 : A mannequin being detected as a person standing at ~ 16 meters from camera
If a person is standing in front of another person in frame then it may happen that one would obscure the view of another person . Hence , only one person will be detected by the system as the camera could not capture both of them .
- 32 - November 2020