FallAction_dataset.zip

FallAction Dataset for Uncertain Action Recognition

Nitika Nigam, Deepali Verma, and Tanima Dutta

Indian Institute of Technology (BHU), India.

Overview:

FallAction is an uncertain action recognition dataset of realistic people falling videos

collected from YouTube. The dataset comprises 15 fall action categories, and each

category contains 50-100 videos. FallAction gives the diversity in terms of different

falling actions and with the presence of noises, such as, variations in camera motions,

person appearance, viewpoint, cluttered background, illumination conditions, etc. It is a

challenging dataset for uncertain action recognition. Most action recognition datasets

are based on certain actions; on the contrary, FallAction aims to encourage further

research into uncertain action recognition by learning and exploring new realistic

uncertain fall action categories.

Structure for FallAction Dataset

● Data associated with each fall action category is stored in separate directories.

● Each directory comprises *.mp4 files for videos.

● The directory is arranged in the following structure:

FallAction_datasets

├── Butt_fall

├── Drunk_fall

├── Elderly_fall

├── Escalator_fall

├── Faint_fall

├── False_start_fall

├── Handstand_fall

├── High_Heel_Break_fall

├── Karate_fall

├── Parkour_fall

├── Pool_fall

├── Side_fall

├── Surfing_fall

├── Trust_fall

└── Yoga_fall