Target Detection Improvements Using Temporal Integrations and Spatial Fusion

摘要

1). Both the pre- and post-detection temporal integrations considerably improve target detection by integrating only 3~5 time frames (tested by real sensor noise as well as computer generated noise). 2). The newly developed pre-detection temporal integration techniques (Additive, Multiplicative, or MIN Fusion) perform a little better than the traditional post-detection temporal integration technique (Persistency Test). Detection results can be further improved by combining both the pre- and post-detection temporal integrations. 3). The techniques developed in this study can be further used for target recognition such as the ATR using matched filtering/correlation approach. 4). The sensor clutter noise looking at real scenes (trees, grass, roads, and buildings, etc.) has been studied. The sensor clutter noise at most (> 95%) of the sensor pixels is near stationary and un-correlated between pixels as well as (almost) un-correlated across time frames. 5). The noise at a few pixels (looking at the grass near the road edge) has shown non-stationary properties (with increasing or decreasing mean across time). 6). Based on observations over the real IR sensor clutter noise, we proposed two advanced thrsholding techniques: the double-thresholding and the reverse-thresholding. They can outperform the traditional CFAR single-thresholding technique when involving complicated clutter situations. 7). From the training data, if we encounter clutters with broader pdf (probability density function) than the target, it would be important to further investigate if the broad clutter pdf is caused by non-stationary noise with a time-variant mean or is caused by a mix of different clutter types with different stationary means. Then we can accordingly select different detection techniques such as the newly proposed double-thresholding or reverse-thresholding schemes discussed in the previous section. 8). It's critical to further investigate and understand the non-stationary noise property under different weather conditions and different background scenes and textures such as grass, trees, sand, and water surfaces, etc.