Human Posture Recognition System

• The objectives of the project are to -
  • Make a dataset for posture recognition.
  • Use the dataset to create a CNN classifier.
• The following postures have been considered -
  • Bending
  • Lying
  • Sitting
  • Standing
• The programs have been written using Python3 in Google Colab.

Dataset

• The dataset is available as downloadable zip file under Dataset folder.

• The size of images available are 256 x 256 and 512 x 512.

• Each of the zip file follow the following directory structure -

  .
  |--Training
  |  |--Bending
  |  |--Sitting
  |  |--Lying
  |  |--Standing
  |
  |--Validation
  |  |--Bending
  |  |--Sitting
  |  |--Lying
  |  |--Standing
  

Structure Description

• Dataset - Contains the dataset.
• ScratchModel - Contains the program used for training the classifier from scratch. It also contains a script to make prediction using the saved model.
• TransferLearning - Contains the pre trained weight files and the programs for performing transfer learning.
• The pre trained model can be downloaded from ./ScratchModel/model/ and executed using ./ScratchModel/model/predict.ipynb