This project is made with a perspective of Recognising Human Activities. The human activity recognition model was trained on Kinetics 400 Dataset. To learn more about this dataset and model used refer Kay et al.’s 2017 paper, The Kinetics Human Action Video Dataset. and Hara et al.’s 2018 CVPR paper, Can Spatiotemporal 3D CNNs Retrace the History of 2D CNNs and ImageNet? The authors Kensho Hara, Hirokatsu Kataoka, Yutaka Satoh have mentioned in their 2018 CVPR paper, how existing state-of-the-art 2D architectures (such as ResNet, ResNeXt, DenseNet, etc.) can be extended to video classification via 3D kernels.

• Python Programming Language
• Convolutional Neural Network

• ONXX-Based Deep Learning models
• OpenCV for Python

• Install Python
  
• Install OpenCV for windows
  
• Install OpenCV for Ubuntu
  
• Execute the following command in terminal:
  pip install numpy
  

• You need to dowwload a resnet-34_kinetics.onnx model from here and once downloaded drop it inside the model directory of our project (shown in directory structure)

1. I have added a video example for testing in test directory
2. If you want to test your own video file be sure to add it in test folder
3. Now, inside recognise_human_activity.py constructor set instance variable VIDEO_PATH to you file path.
4. Otherwise, if you want test the model on using web-camera live video just set self.VIDEO_PATH = None

• Once your setup is done run the following to execute code:

python recognise_human_activity.py

• Chaitanya Sonavane [July 2020]

• Kay et al.’s 2017 paper, The Kinetics Human Action Video Dataset.
• Hara et al.’s 2018 CVPR paper, Can Spatiotemporal 3D CNNs Retrace the History of 2D CNNs and ImageNet?
• OpenCV's Action Recognition Example