Python 3.7 Tensorflow 1.15 LabelImg
There are endless applications of AI, My thought process was to apply AI in Construction Industry in a way that could help improve construction workers' safety. This application of the model to CCTV Video can add as a safeguard to ensuring the safety of workers in the field.
I collected about 2K images from Google and labeled these images using 'LabelImg'. labelImg is a tool in python that has a straightforward interface to label our classes. In this case, 'Helmet_on' and 'Helmet_missing'. These images are then converted to a TensorFlow record for training. We are using the Tensorflow Object detection API here.
Since I cannot train a model from scratch with limited images, I will apply Transfer learning here and use Faster-r-CNN-resnet pre-trained model weights as my starting point and modify it to detect classes for this specific use case.
The model is trained until loss < 0.05
https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/training.html