google2c7a87877ffe6fb7.html

Video-Based Abnormal Driving Behaviour Detection via Deep Learning Fusions

if you want the project pls call @8125424511

Video-Based Abnormal Driving Behaviour Detection via Deep Learning Fusions

In this paper author is describing concept to detect abnormal driving behaviour from videos using Deep Learning Algorithms such as Wide Group Densely (WGD) Network, Wide Group Residual Densely (WGRD) Network and Alternative Wide Group Residual Densely (AWGRD) Network. All algorithms build training models to detect abnormal behaviour but AWGRD works better than other two algorithms so I implemented AWGRD Algorithm.

Existing CNN Algorithm works better by monitoring eyes, mouth, monitor heart by using sensor and monitoring hands behaviour by using sensors but this algorithms give false prediction. For example while eyes monitoring if user turn around then algorithm will not detect eyes and consider driver is sleeping and sometime driver go to sleep without closing eyes, Using sensors for monitoring will put extra burden on driver.

To overcome from above issues author has describe 3 algorithms based on CNN deep learning models.

1) Wide Group Densely (WGD) Network:Technically,WGDtakes important issues of deep learning models, i.e., thedepth, the width and the cardinality, into consideration whendesigning its model structure based on Dense Net. This model use deep features from input train model to get better prediction accuracy.

2) Wide Group Residual Densely (WGRD) Network: Themost significant change of WGRD with respect to WGD isthat, the idea of residual networks is incorporated in WGRD. In this algorithm input image will pass from one layer to other residual layer to have best features from train input image to get best accuracy.

3) Alternative Wide Group Residual Densely (AWGRD) Network: This algorithm works similar to above two algorithms but while passing input data from one layer to other, this algorithm will take super positions of previous layers which has best features from all layer and will have better prediction accuracy. Due to super positions extraction training efficiency willundoubtedly become higher.

Module Information

1) Generate & Load AWGRD Model: Using this module AWGRD train model will be generated from input images download from Kaggle state farmdistracted driver detection database. This database contains 22424 images and model is built by using all those images.

2) Upload Video: using this module we can upload video to this application and then start playing video using Python OPENCV library.

3) Start Behaviour Monitoring: Using this module we will extract each frame from video and then resize image according to AWGRD Model. AWGRD Model will be applied on this frame to predict behaviour of driving person. All behaviours will be displayed on playing video.