Testing apparatus and method for special object

Abstract

This invention relates to a method and device to catch object from multiple point, which divides special capture space into multiple sub special capture spaces and divides the sub special capture space into multiple spaces from multiple visional point and comprises the following steps: establishing tree shape tester with one large branch; the branch covers all sub space with multiple branches with each one responding to one branch; adopting vector induction formula to train each branch into determine vector to determine each point to those down layer point; testing the sent point object as special object with relative visional angle.

Description

technical field [0001] The present invention relates to an effective technology applied to a device and method for detecting a specific subject or a part of a subject, such as a person, an animal, or an object, contained in a captured image. Background technique [0002] As a conventional technique, there is a technique of detecting a specific subject or a part of a subject, such as a person, an animal, or an object, contained in a captured image. As an example of such a conventional technique, there is a technique of detecting the face of a person from a captured image, that is, a face detection technique. [0003] Face detection has two difficulties: one is the internal changes of the face, such as the difference in face shape; the other is the external changes of the face, such as rotation in the plane. Therefore, only by finding some related algorithms and achieving real-time application effects can a face detection device with practical application value be constructed...

AI Technical Summary

Problems solved by technology

Such a classification problem is too difficult, and it needs to construct a more complex classifier to solve it. Therefore, although good detection accuracy has been achieved, the speed is not ideal.

On the other hand, the Decision Tree structure emphasizes the differences between faces of different poses. First, a classifier is constructed to classify them, and then they are separated from non-faces. The speed of this method is ideal, but the accuracy And the robustness is not ideal, this is because the multi-classification of face pose itself is a problem that is not easy to solve, and this method hopes to use the decision tree to quickly determine the face pose will inevitably lose a certain degree of accuracy and practical performance

In general, the coexistence of consistency and difference between faces in different poses is a key problem that cannot be overcome by the above two methods.

[0012] In fact, for face images in real life, RIP and ROP are common and exist at the same time, and the existing methods cannot meet the requirements of practical applications, such as the need for rotation invariance of MVFD

Although these problems can be solved by rotating the image and applying MVFD several times, it increases the computational complexity and false positives

Images