Master thesis presentation by Wenjing Chen, MPBME
När: 27/10/2017 , 14:00
Plats: Lunnerummet (room 3311), 3rd floor
Adress: Chalmers Johanneberg , Göteborg
On Friday, 27 October, Wenjing Chen, MPBME, will present the master’s thesis with the title “Deep Learning for Human Fall Classification with Application to E-Healthcare”.
When? 14:00 pm. on Friday, 27 October, 2017
Where? Lunnerummet (room 3311), 3rd floor, Chalmers Johanneberg, Göteborg
Examiner: Irene Gu
With the increasing of computation power and amount of data, deep learning has achieved a great success in computer vision. Human activity recognition is one of the important tasks in computer vision. Fall detection/classification is an import issue in e-healthcare and assisted-living. This thesis focuses on using deep learning for classification of human falls from videos. In the first part of the thesis, image classification using CNN is investigated on traffic sign images, to study the network learning process. In the second part, i.e., the main part of the thesis, fall detection/classification is studied with three different network structures, namely, single-frame-based 2DCNN, multi-frame-based 2DCNN, and RNN. Here, single-frame-based 2DCNN is used as the baseline, while multi-frame-based 2DCNN and RNN are then experimented for utilizing temporal information.
Experimental results from this thesis work will be presented. For the GTSRB traffic sign dataset, a high test accuracy of 98.83% is obtained. For the multiple camera fall dataset, the test accuracy of 85.42% is obtained.