A multi-target object detection method and device

Abstract

The invention discloses a multi-target object detection method and device. The method comprises: connecting a target object detection device; Creating a pre-trained multi-target object detection modelby using the convolutional neural network; Installing deep learning framework software; Sequentially reading each frame of image from the camera; Reducing the image read by the camera to 448 * 448 pixels; Dividing the shrunk image into 7 * 7 grids with the same size; Judging whether the broken object is in a 7 * 7 grid unit or not by using the coordinate value; Sending the grid unit with the object into a pre-training network model to obtain a frame regression value; Outputting the frame regression of the 90 object categories of each grid; Outputting the position value and the confidence degree of each border regression object; Setting a threshold value to filter out frames with low scores; And performing non-maximum suppression processing on the reserved frames, and combining the framesto obtain a final detection result. The method solves the problems that in the prior art, the design of image feature extraction is tedious, the detection speed is low, and the multi-target concurrency capability is poor.

Description

technical field [0001] The invention belongs to the technical field of computer image processing and machine vision, and relates to a multi-target object detection method, and more specifically relates to a two-dimensional video camera multi-target object detection method and device. Background technique [0002] Traditional object detection generally uses a sliding window framework, which mainly includes three steps: (1) using sliding windows of different sizes to frame a certain part of the image as a candidate area; (2) extracting visual features related to the candidate area. For example, the Harr feature commonly used in face detection; the HOG feature commonly used in pedestrian detection and ordinary target detection; (3) using classifiers for recognition, such as the commonly used SVM model. In traditional target detection, the multi-scale deformable part model DPM regards objects as multiple components (such as the nose and mouth of a human face, etc.), and uses the...

AI Technical Summary

Problems solved by technology

This implementation requires high computing resources, and it is difficult to handle multiple targets at the same time

When detecting multiple targets in a real-time camera video stream, accelerated training with multiple GPU graphics cards is often required, making the device for target object detection very poor in portability

It is often difficult to apply in some end-to-end real-time processing fields without network and high mobility requirements

Images