According to the real-time non-tracking monitoring video remnant detection method provided by the invention, manual
feature extraction and
deep learning recognition are combined, and non-tracking remnant detection is realized. The method comprises the following steps: firstly, on the basis of a
frame difference method, counting continuous
frame sequence change conditions of a foreground region toobtain an initial static target region; Then, we will make Two kinds of manual
design characteristics of a
gradient direction straight directional diagram and
hue- Saturation-
lightness combineto carryout suspicious object pre-judgment, and a
pseudo static target area caused by influences of illumination changes and the like is eliminated. And finally, known objects and pedestrians are excluded bycombining a
deep learning technology, so that final confirmation is carried out on suspicious objects, and non-tracking detection of the remaining objects is realized. According to the method, pseudotargets generated by illumination change and
pedestrian retention in a scene can be eliminated, remaining objects can be accurately detected, the single-frame
processing time is shorter than that ofother two methods, and the requirement for real-time alarming can be met.