PV056-Anomaly-detection-from-camera-data

The task of this project is to find suitable and accurate way of detecting abnormal camera view on given image data using several methods of machine learning. Data are collected from several camera sources from different places, and for simplicity chosen only from daytime. Given image data displays a surveillance pictures of a road and its surroundings, with vehicles passing by. The datasets consist of normal images, and images with some type of abnormal camera view. There are different types of anomalies displayed on the pictures, whether the anomalies are caused by camera malfunctioning or transmission error, or by external causes (vehicle blocking camera view, objects on road, people on road etc.). The goal of the project is to find a successful method for detecting various abnormal camera views on given dataset using following methods of machine learning: 1) Autoencoders and neural networks 2) SVM and bag of visual words Part of the solution should be comparison of the results of the named methods of machine learning and also comparison of their advantages and disadvantages on given data.