Fall recognition system using convolutional neural network
Tsepo Kolobe, Chunling Tu, Pius Adewale Owolawi · 2022 International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME) · 2022
Fall imposes a great challenge not only to our families but also to the economy of our countries. Due to fall, there is a high mortality rate of elderly and disabled people. Modern technologies are utilized to provide secure and safe living environments. Although the use of computer vision and deep learning techniques provides an interesting solution to mitigate fall, they still experience overfitting problem due to scarcity of fall data. In this proposed system, we work towards improving the fall recognition accuracy of convolutional neural network (CNN) by performing hyperparameter tuning and applying data augmentation techniques. The experiment shows that the proposed model achieves interesting results with an accuracy of 99.8%. It is clear that CNN approach provides the optimal solution for identifying falls from activities of daily living.