Spatio-Temporal Pattern Recognition Using Hidden Markov Models.

摘要

A new spatio-temporal method for identifying 3D objects found in 2D image sequences is presented. The Hidden Markov Model technique is used as a spatio-temporal classification algorithm to identify 3D objects by the temporal changes in observed shape features. A new information theoretic argument is developed that proves identifying objects based on image sequences can lead to higher classification accuracies than single look methods. A new distance measure is proposed that analyzes the performance of Hidden Markov Models in a multi-class pattern recognition problem. A three class problem identifying moving light display objects provides experimental verification of the sequence processing argument. Individual frames of a MLD image sequence contain very little spatial information. The single look classification rate for the moving light display imagery was observed to be near 50%. In contrast, the Hidden Markov Model classification rate was above 93 %. The alternate nearest neighbor multiple frame technique classification rate was 20% below the Hidden Markov Models. A one sided t-test revealed a highly statistically significant difference between the Hidden Markov Model and multiple frame technique at a 0.01 level of significance. A five class problem consisting of tactical military ground vehicles is considered to provide verification using imagery with both spatial and temporal information. Results confirmed the new spatio-temporal pattern recognition method produces superior results by accessing the temporal information in the image sequences. A prototype automatic target recognition system is demonstrated. Hidden Markov model, Pattern recognition, Motion analysis, Signal processing.