Abnormal detection refers to infrequent data instances that come from a diverse cluster or distribution than the majority normal instances. Owing to the increasing demand for safety and security, discovery abnormalities from video streams has attracted significant research interest during recent years. The current advancements in computer vision and deep learning have a remarkable role in enabling such intelligent frameworks. Different algorithms that are specially designed for building smart vision frameworks seek to scene understanding and building correct semantic inference from observed dynamic motions caused by moving targets. In this project the used models is Fast R-CNN for human detection and SSD for pose detection. Unfortunately, although there are many algorithms have been proposed in this interesting topic, the research in this area still lacks strongly to two important things: comparative general assessment and public-accessible datasets. This project is mainly focus on the detection of abnormal human behaviour in restricted environment.
Refer and clone the following repositories: https://github.com/ildoonet/tf-pose-estimation and https://github.com/tensorflow/models .
The project is done under LeadingIndia.AI Inernship.
Mentor of the project: Shreyans Jain and Dr. Kanad Kishor Biswas
For Complete details please read the Report.