Facial Recognition is tricky in real life scenarios, where in angle of face and illumination , blurred images etc. lead to many false positives. We have worked on many FR models and we don’t claim to be the best in the market, however, we have solved many complex problems of blurred images and even rotated faces, we have worked on following models of FR:
backends = [‘opencv’, ‘ssd’, ‘dlib’, ‘mtcnn’]
models = [“VGG-Face”, “Facenet”, “Facenet512”, “OpenFace”, “DeepFace”, “DeepID”, “ArcFace”, “Dlib”]
In most cases the video feed from camera comes from 3MP cameras , this leads to many issues in real life , you start getting lot of false positives while this can’t be removed completely but we have achieved good results in hostile conditions as well as show in images below:
Our Algorithm pulls out true detection even when faces are occluded to a certain degree.
