Abstract:
In large scale visual surveillance applications, classification
of human behaviors is very important. Classes of interest include
suspicious human behaviors which should be effectively
detected so as to alert supervisors' attention. In this paper,
a data-based neural network such as the Modified Probabilistic
Neural Network (MPNN) is introduced to approximately
partition the classification space nonlinearly in order
to achieve an acceptable classification performance while reducing
computational complexity. The paper shows that this
kind of network is able to achieve a good trade-off between
classification accuracy and computational complexity. The
performance of MPNN is compared to that of more conventional
classification methods such as Hidden Markov Models
(HMM) and the Multilayer Perceptron (MLP).
