Target detection is an important direction in the field of computer vision, and its main task is to locate and classify target objects in images or videos. According to the algorithm flow, the target detection algorithm can be divided into two genres: one is the Two-Stage algorithm, which detects the target is mainly divided into two parts, 1) through a special module to generate candidate frames, find the foreground and adjust the bounding box; 2 ) to classify and regress the candidate boxes. The other is the One-Stage algorithm, which directly performs dense sampling on the entire image, and then directly performs classification and regression after using CNN to extract features.

 

In addition, before deep learning intervened in this field, traditional object detection methods included Viola-Jones and HOG+SVM, etc. Viola-Jones uses integral image features + AdaBoost method to detect faces, etc.; HOG+SVM is mainly used for pedestrian detection, by extracting HOG features from pedestrian target candidate areas, and combining with SVM classifier for judgment.

 

These are just some classic approaches in the field of object detection, and there are many others to choose from.