Anchor-frame-free target detection method and system based on diagonal network

Abstract

The invention provides a diagonal network-based anchor-frame-free target detection method and system. The method comprises the following steps of: extracting depth features of an input image by using an hourglass network and carrying out pooling operation so as to obtain a key point thermodynamic diagram, an embedded feature vector feature diagram and an offset feature diagram; processing the key point thermodynamic diagram by using a diagonal network, sorting the key points according to confidence coefficients, selecting the key points of the first k confidence coefficients as target key points, and obtaining key point positions and key point category information of the upper left corner, the lower right corner and the central point; based on a Fisher criterion, designing an embedded vector learning loss function to carry out key point combination pairing training, and further discriminating a preliminary pairing result by using a central key point; and using the distance, the length and the slope measurement loss of the diagonal center points of the prediction rectangle and the annotation rectangle frame to design the diagonal loss for regression training of the target prediction frame. The technical problems that the anchor frame is difficult to design, samples are unbalanced and the detection precision is low are solved.

Description

technical field

[0001] The invention relates to the field of computer vision, in particular to a method and system for anchor-free frame target detection based on a diagonal network. Background technique

[0002] Object detection is an important research direction in the field of computer vision, and its purpose is to predict the location and class of the bounding box of each object of interest in an image. Object detection has a wide range of applications in the fields of autonomous driving, video surveillance, visual guidance, and air defense early warning. At present, target detection based on deep learning mainly includes two categories: anchor-based target detection method and anchor-free target detection method.

[0003] In the early stage of research, the target detection method based on anchor frame has received extensive attention, which has significantly improved the precision and recall rate of target detection, but it still has two problems: First, the layout of...

Images