ACTIVITY RECOGNITION METHOD BASED ON A SEMI MARKOV CONDITIONAL RANDOM FIELD MODEL, PARTICULARLY FOR SIMULTANEOUSLY PROGRESSING TRAINING AND DEDUCTION IN A SEMI MARKOV CONDITIONAL RANDOM FIELD MODEL

摘要

PURPOSE: An activity recognition method based on a semi Markov conditional random field model is provided to efficiently capture behavior variations during long time period by simultaneously training and detection in a semi Markov conditional random field model.;CONSTITUTION: An input signal measured by an accelerometer is split and outputted as frame sequences(S31,S32). An eigenvector is extracted from the frame sequence, and features are extracted by using a training input signal as one frame set unit and collected as a kernel vector(S33). The eigenvector is quantized based on the kernel vector most similar to the eigenvector and outputted as discrete input sequence. A semi Markov conditional random field model receives the discrete input sequence and calculates the probability of the state sequence.;COPYRIGHT KIPO 2011