Moving target indication (MTI) methodologies are essential tools to detect, locate, recognize, and track the moving targets of interest in a movie or video sequence. Different MTI methodologies can be characterized and compared by their hit rates (percentage of all available targets of interest detected and located), recognition rates (percentage of each of the existing target types correctly recognized), false-alarm rates (average number of false alarms reported per video frame), efficiency of operation (how much computational resources and time is needed for a given set of data), and robustness (how well the methodology is able to handle or adapt to different types of data). An ideal MTI methodology should generally be able to detect, recognize, and track all targets of interest without incurring an unacceptable number of false alarms under a very stringent computational requirement.
Generally, most conventional MTI methodologies look for changes in a video sequence by subtracting the current image frame (arrays of digital pixels) being viewed from the previous one. While it is the simplest way to do so, this method typically produces more false alarms and generally does not work well when the targets are moving slowly relative to their sizes. Additionally, problems in variations in contrast, brightness, and other video parameters — as well as aliasing, jitter, and background errors — can cause positive false alarms.
Some conventional techniques attempt to stabilize the background information and reduce the false alarms by creating an average image consisting of several previous frames. However, these conventional methods tend to create a trailing ghost shadow that causes new false alarms, and that are generally difficult to suppress, especially when some of the moving targets are brighter than their surroundings while others are darker than their background in the same video sequence. In this situation, conventional methodologies tend to result in either detecting only those targets in the chosen polarity and the shadows of the opposite polarity, or the targets and shadows of both polarities. If half of the moving targets are brighter than their surroundings and the other half is darker, then either nearly half of the targets would be forsaken or nearly twice as many false alarms would be generated. Obviously, neither one of these two cases is acceptable to a robust tracking methodology.
Unfortunately, the conventional solutions have generally not been able to overcome these shortcomings. Therefore, there remains a need for a novel MTI methodology that is capable of detecting, recognizing, and tracking most, if not all, of the interested targets with an acceptable number of false alarms under a very stringent computational requirement.
The unique software solution invented in this work uses adaptive background models that enable detection of legitimate targets in a busy scene, subtracting those targets from less interesting background changes, and creating an accurate change map, thus reducing human error and computergenerated false positives. The algorithm is robust and is not susceptible to reverse polarity where the brightness or intensity of a moving object changes throughout a video sequence. With this method, once the target is identified, it remains clearly distinct from the background. The method is also data-efficient, reducing computational requirements as compared to competing methods. While initially developed for forward-looking infrared images, the algorithm is adaptable to most types of video imaging.