Use of probability hypothesis density filter for human activity recognition

Günay, Elif Erdem
This thesis addresses a Gaussian Mixture Probability Hypothesis Density (GMPHD) based probabilistic group tracking approach to human action recognition problem. First of all, feature set of the video images denoted as observations are obtained by applying Harris Corner Detector(HCD) technique following a GMPHD lter, which is a state-of-the-art target tracking method. Discriminative information is extracted from the output of the GM-PHD lter and using these, recognition features are constructed related to di erent body segments and the whole body. An unique Hidden Markov Model(HMM) belonging to each feature is fed by these information and recognition is performed by selecting optimal HMM's. The performance of the proposed approach is shown on the videos in KTH Research Project Database and custom videos including occlusion scenarios.The results are presented as the percentage of the correctly recognized videos. Same experiments on KTH database are performed for KLT tracker instead of GMPHD in the proposed approach. In addition, a comparison is made for an algorithm in the literature for the custom videos. The results shown that proposed approach has comparable performance on KTH database and is better in handling occlusion scenarios.