Head detection method and device

Abstract

An embodiment of the present invention provides a method and device for head detection. The method includes: inputting an image to be tested into a head detection model, obtaining a set of bounding boxes, the classification confidence of each bounding box in the set of bounding boxes, and each The IOU positioning reliability of the bounding box; based on the classification confidence of each bounding box in the bounding box set and the IOU positioning reliability of each bounding box, the IOU-NMS algorithm is used to carry out back-end processing on the bounding box set, A head detection result is obtained; wherein, the head detection model is obtained after training according to the image samples and the head target truth boxes corresponding to the image samples. In the embodiment of the present invention, the head detection model composed of a convolutional neural network and a specific cyclic neural network model is used to perform feature extraction and decoding regression on the input image, and then use the IOU-NMS algorithm to perform back-end processing on the bounding box, which can effectively improve complex scenes. The accuracy of head detection.

Description

technical field [0001] The invention relates to the technical field of target detection, in particular to a head detection method and device. Background technique [0002] Target detection is to locate the location area where the target object appears in the image and determine the classification of the target object. With the rapid development of computer vision and pattern recognition technology, as well as the continuous improvement of hardware performance, object detection technology has been widely used. In large public places such as schools, shopping malls, and hospitals, crowd activities are frequent and dense. Detecting the characteristics of people in these complex scenes to obtain information plays an important role in public security prevention and control, rational deployment of public facilities, and business information collection. [0003] The selected target of human body detection can be a complete human body, or a complete head, face, limbs and other loca...

AI Technical Summary

Problems solved by technology

However, the current methods based on regression ideas are not accurate enough for head detection in complex scenes with varying degrees of occlusion, deformation, rotation, and scale changes, and cannot meet the application requirements.

Images