Overview / Usage

1)Real time detection of basic activities like running, jogging, walking, handwaving... The project can be extended to cctv surveillance and object tracking in input video..

Methodology / Approach

First step is to extract the frames from the videos and make prediction on each frame using inception-v3 model.Since Transfer Learning technique is used the final classification of inception is not extracted instead extract the result of last polling layer which is vector of 2048 values.Repeat for all the frames corresponding to the respective videos.pass the frames through sequential model (LSTM) to obtain final classification of the system.

Technologies Used

Opencv

PIllow

Keras ,matplotlib,tensorflow,numpy,sk-learn

sk-video

ffpmpeg

GPU - NVIDIA