PhD Topic : "Machine learning method for Knowledge Extraction from the video "

Abstract :

We have essentially developed a event detection system from the video. For this, we tried to see the various descriptors of the video starting with static descriptors. These descriptors can be applied either directly on the image / scene, or indirectly by applying them to regions or points of interest or segments. These descriptors can also be merged (with certain precautions) to construct a descriptor more efficient to characterize the image / scene more precisely. It was later found, the appropriate descriptor for each concept in evaluating these descriptors by the concept research using SVM as the classification technique.
Then we extracted spatio-temporal descriptors to characterize the video to detect events. Were calculated descriptors based on the movement of image blocks to model events. We developed a system based on the HMMs modeling.
Our job will be to develop a method of feature extraction to detect events in video that will be efficient and quick to achieve good results that we can compete in the competition TRECVID.

Keywords : Knowledge extraction, machine learning, Hidden Markov Models, processing and analysis of the video, Visual content based Event detection in video