Description: Anomaly Detection Using Deep Models
What is an anomaly?
True definition of anomaly is "different from usual/normal". Subjective to this project, it also means something similar. An accident, a robbery, a carjacking, a street fight and many such illegal activities are considered an anomaly according to this project.
Main aim of this project is to detect and describe above mentioned anomalies.
First manifestation of this project is a deep learning image detection model with classes as the anomalies.
Later manifestations of this project is a deep learning sequence model.
As the project requires a model to be developed most probably a deep learning model, data is required, loads and loads of data is required. Present idea to collect data, is to gather all open source crime datasets, accident datasets, and all illegal activity datasets. At the intial stages the model is most probably going to be image model, so the datasets are image datasets. Later extending to video datasets.
Later on the idea is to train the model on the CCTV footages, which requires a lot of labelling of the data.
If this model is introduced into CCTV cameras, and real time detection is made, could help save some lives and property. Upon detection of an anomaly, concerned authorities are informed automatically. Help reaches where it is required the fastest. Also, decreases human labour which can be not so accurate at times of mental pressure.
Also, the traffic regulations detection must be inculcated in it. One stop solution to all the problems.
Additionally, audio feed is also considered for training to improve the accuracy.