Small target detection in wide-area surveillance is a challenging task. Current imaging staring sensors in practical systems are characterized by large pixel counts and wide field of view (WFOV). Therefore, it is not suitable to detect targets simply via a single algorithm in different background types. We solve this problem by local windowing approach and take different background suppression and target enhancement methods for different surveillance scenarios. Monte Carlo simulations are provided and the experimental results demonstrate that we can effectively detect dim small targets with a very low false alarm rate and an acceptable detection rate in the proposed detection architecture.
In order to detect the dim small target in high frame rate image sequences，an optimized temporal processing technique is
investigated. Based on the temporal profile models for noise pixel，target pixel and clutter pixel, we formulate the
detection in two steps, pre-processed by Max-median filter and temporal variance filter(TVF). In pre-processing step,
three spatial-filtering methods are compared. In temporal profile analysis step, the length of time windows for calculating
the mean and variance values are chosen after statistical analysis. Finally, six targets embedded into a scene which
contains different types of clouds, and set the adjacent scene to one pixel jitter in any random direction. The simulation
results show that we can obtain a relative high signal-to-clutter gain in different regions, which satisfies the requirement
of target detection algorithm in high frame rate detection system.