SeVIA - Automatic semantic-supported video analysis for new media services

In this project it should be examined how computer vision methodes can be used to realize an innovative and internationally marketable product for the company Future TV. The aim is to build a new software that can automatically generate a semantically-supported description of the content present in a video sequence. A wide spektrum of relevant objects (e.g. car, TV, chair, person...) should be detected in the video and used then for example to generate a klickable bounding box around the object. Additionally, the scene in the video should be classified in the proper category (e.g. beach, bar, street, football stadium...) and the genre shoud be recognized (news, sport event, Crime scene...). With the help of that description, promotional videos can be optimally selected and recommended for the users. Convolutional neural networks are the basic technology that is used in this project. The used object detection models and hierarical image classification models are all CNN-based.