Motion detection method recognizes the target by obtaining the target motion information, and the most commonly used method is to analyze the gait of the candidate target. Due to the need for temporal information, this method cannot identify the human body in stationary and unconventional gaits, and requires visible legs or feet.

 

It can be seen from the above analysis that any single method has certain deficiencies in infrared human target recognition. The performance of the human body model method is generally not high, and the practicability is poor. The overall template matching requires a perfect template set, but it is difficult to achieve a perfect template set, and there must be a strategy for fast matching with good performance. Part-based template matching has some advantages in dealing with complex poses and occlusions, but how to integrate multi-part information needs further research. The core problem of statistical classification is to find effective human features and in-depth research on the structure and performance of classifiers. The motion detection method has mature algorithms available, but it is ineffective for stationary targets, and the definition of the motion period is also ambiguous, so the corresponding detection accuracy is not easy to improve.