Overview / Usage

In recent years, Automatic fall detection using Human Action Recognition has received huge attention amongst the researchers because falls are one of the major threat to the quality of life of people especially elderly people and a serious fall at home or at workplace, may lead to severe injuries and even death if person cannot call for help and remains on the ground for too long.
Some methods are generally the ones which uses sensors like vibration sensors in order to detect falls. These sensor based methods even provide high detection rates but still some of the people are reluctant to wear them at all times and these sensor based methods even limit the performance by increasing false detection rates and thus their use is not standardized. As a consequence, another approaches focusing on vision based methods, came into light. One of the biggest advantages of these vision based fall detection is that it doesn’t require anyone to wear anything. In addition, vision based systems provide more information about the motion and behavior of the person than the sensor-based systems.

Methodology / Approach

Transfer Learning approach for Human Action Recognition is achieved by using Xception model for feature extraction along with a classifier for action recognition. The reason behind using the Xception model is because it sports the smallest weight serialization.

Technologies Used

Python
OpenCV
CNN
Keras