Skip to content

Final Year Project: Implementation of deep neural network for people identification from video by the characteristic of their gait.

Notifications You must be signed in to change notification settings

umairarshadbutt/Blue-Security

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 

Repository files navigation

Blue Security

hire-badge hire-badge

In this project you can find implementation of deep neural network for people identification from video by the characteristic of their gait. The processing is very robust against various covariate factors such as clothing, carrying conditions, shoe types and so on.

Wana know more? Watch the presentation video.

screenshot

Requirements

Software Requirements

• TensorFlow 1.8
• Cuda 9.0
• CuDNN 7.0
• numpy, scipy, PIL
• python 3.5.2
• Tknr

Hardware Requirements

• RAM 4GB
• GPU 11GB
• Architecture x64
• IP camera
• WebCam

Basic information about architecture

The network takes raw RGB video frames of a pedestrian as an input and produces one-dimensional vector - gait descriptor that exposes as an identification vector. The identification vectors from gaits of each two different people should be linearly separable. Whole network consists of two sub-networks connected in cascade – HumanPoseNN and GaitNN. Spatial features from the video frames are extracted according to the descriptors that involve pose of the pedestrian. These descriptors are generated from the first sub-network - HumanPoseNN defined in human_pose_nn module. HumanPoseNN can be also used as a standalone network for regular 2D pose estimation problem from still images. Responsibility of the second sub-network - GaitNN defined in gait_nn module is the further processing of the generated spatial features into one-dimensional pose descriptors with the use of a residual convolutional network. Temporal features are then extracted across these pose descriptors with the use of the multilayer recurrent cells - LSTM or GRU. All temporal features are finally aggregated with Average temporal pooling into one-dimensional identification vector with good discriminatory properties. As already mentioned in the text above, the human identification vectors are linearly separable with each other and can therefore be classified with e.g. linear SVM.

Author

👤 Umair Arshad

👤 Junaid Hussain

👤 Muhi O Deen

About

Final Year Project: Implementation of deep neural network for people identification from video by the characteristic of their gait.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published