To segment a sequence of images whose background may be static or may change rapidly, there are two common processing methods: motion detection method and static detection method.

 

The motion detection method uses the time domain information to segment the motion area in the sequence image, and the typical representatives are the frame difference method and the optical flow method. The frame difference method uses the difference between frames to achieve moving target detection, which has the advantages of simplicity and real-time performance. However, the detected moving targets are usually incomplete and require more post-processing to form a meaningful target area. The optical flow method realizes moving target detection through optical flow calculation, and has strong adaptability to background changes, but the calculation amount is quite large and the real-time performance is poor.

The static detection method ignores the motion information in the image, regards the sequence image as a combination of a series of independent static images, and then uses threshold comparison, template matching and other means to segment the image target. The static detection method does not need to assume the background state of the sequence image, but due to the lack of utilization of motion information, the detection results are prone to false foreground and target fragmentation problems.

This section mainly describes a segmentation method combining motion detection and static detection, which achieves a good segmentation of human targets in thermal infrared sequence images with the above-mentioned complex background changes. Because there is no need to assume the background state, and the adaptability to background changes is strong, it can meet the needs of the moving infrared human target detection system. In this method, both motion detection and static detection functions are implemented by the LSAC model. Among them, the motion detection LSAC model combines the basic idea of background subtraction, which can realize foreground detection and background update at the same time. The static detection LSAC model uses a double-threshold technique to detect image areas within a specified grayscale range. The above LSAC model is applied to the sequence image, and after the motion detection area and the static detection area are obtained respectively, the segmentation result is obtained through the fusion process. This greatly reduces noise in segmentation results and improves human target integrity. In addition, the fast numerical algorithm is used to solve the LSAC module and the algorithm flow is optimized, so that the method has high operating efficiency.