The basic principle of the detection algorithm is to learn a classifier through the training set, and then slide scan the whole image with different scale windows in the test image; do a classification each time, and judge whether the current window is the target to be detected. The core of detection algorithm is classification. The core of classification is what features to use and which classifier to use.
在计算机视觉领域，最基本也最经典的一个问题就是目标识别给出一张图像，用 detector 检测出图像中特定的 object(如人脸)。检测算法的基本原理就是先通过训练集学习一个分类器，然后在测试图像中以不同 scale 的窗口滑动扫描整个图像；每次扫描做一下分类，判断一下当前的这个窗口是否为要检测的目标。检测算法的核心是分类，分类的核心一个是用什么特征，一个是用哪种分类器。
In the field of computer vision, the most basic and classic problem is to give an image by object recognition, and use the detector to detect specific objects (such as face) in the image. The basic principle of the detection algorithm is to learn a classifier through the training set, and then slide scan the whole image with different scale windows in the test image; do a classification each time, and judge whether the current window is the target to be detected. The core of detection algorithm is classification. The core of classification is what features to use and which classifier to use.
About 80% of the external information that human contact belongs to visual information. For human beings, image and video are the vivid description of objective things and the main source of human information. Target detection and tracking is a hot topic in the field of computer vision. It combines the advanced technologies of image processing, pattern recognition, artificial intelligence, automatic control and many other fields. It has been widely used in intelligent transportation system, intelligent monitoring system, industrial detection, aerospace and many other fields.
Due to the variability of objects (especially pedestrians) and scenes in the real world, it is difficult to use a consensual method to study them. The main problems of target detection are: how to segment the target accurately and quickly, how to minimize the influence of complex background on target detection, and how to reduce the decline of target detection accuracy caused by the change of target scale, size and shape. In addition, in the target detection system, there is a contradiction between the robustness and real-time performance of the system.
The research of object detection mainly includes object detection based on video image and object detection based on static image. This paper mainly discusses the target detection algorithm based on the static image, that is to detect and locate the set type of target in the static image. The difficulty of object detection based on static image is that the object in the image will change [} 2} because of the change of illumination, angle of view and the inside of the object. In view of the above difficulties, scholars at home and abroad have made many attempts. The proposed methods are mainly divided into shape contour based object detection algorithm and feature-based object detection method.
Detection algorithms can be divided into six categories, namely, inter frame difference method, background modeling method, point detection method, image segmentation method, clustering analysis method and motion vector field method. Among them, frame difference method and background modeling method are the most commonly used and simplest algorithms, and have achieved good results in the research, but the two methods have a common feature that they are only suitable for moving object detection in the case of static background.