The statistical identification of motion anomalies in ground moving target radar imagery

The manual analysis of moving target indicator (MTI) imagery is a complex and time consuming process. Therefore, automated methods of statistically identifying anomalous patterns within data in real time are important. Applications include the evaluation of ground surveillance scenery to locate lost individuals in the presence of wind-driven debris, the identification of traffic patterns and foliage locations. Additional use can be made of automated techniques on board space probes that are evaluating atmospheric and terrain disturbances on distant planets. As the volume of data grows and as the time criticality of the data products becomes more urgent, the importance of this automated analysis grows. In our paper we apply multiscale vector quantization to ground-based range-Doppler radar imagery. The vector quantization codebook that results is used to segment the background into regions of statistically similar backscatter range and Doppler variation. Within these regions, smaller scale patterns are used to identify anomalies associated with non-naturally occurring objects. The time progression of the anomaly location is further used to enhance the feature set that is used to ascribe a class to each of the anomalies. We introduce the viewer to the Applied Radar radar testbed data collection system that produced the data used in this study using block diagram and processing descriptions. We then present the GMTI imagery that was used for our study. Within that data we highlight the identified anomalies and give the feature set associated with each. These anomalies are produced by individuals, vehicles and naturally present foliage. Preliminary classifier results are then presented including spectrogram and color mapped images of the data, region feature tables, process flow charts and image examples of the vector codebook members that result. This paper is of interest to researchers involved in analyzing any kind of multidimensional time-varying scenes.