A fast target detection method for binocular images based on two-stream convolutional neural network

Abstract

The invention discloses a binocular image rapid target detection method based on a dual-stream convolutional neural network, comprising the steps of: calibrating a binocular camera to obtain calibration parameters; correcting a training image according to the calibration parameters, and training implicit depth semantic mining The network is used to implicitly learn deep semantic information on binocular images and train a multi-modal feature hybrid detection network; the features output by the implicit deep semantic mining network and the features of the multi-modal feature hybrid detection network are connected in series through channels Combined together, a dual-stream convolutional neural network is formed, and the training image is used to train the dual-stream convolutional neural network; the test image is obtained through the binocular camera, and it is corrected, and the corrected image is input into the above-mentioned dual-stream convolutional neural network Perform target detection to obtain target detection results. The present invention can comprehensively utilize the complementarity of RGB and depth semantic information, and has the advantages of high efficiency and more accurate target detection results.

Description

technical field [0001] The invention relates to the research field of target detection in video surveillance, in particular to a fast target detection method for binocular images based on a dual-stream convolutional neural network. Background technique [0002] The task of target detection is to determine the position of the target object of interest in the image and identify its category. Object detection is the basis for many advanced computer vision tasks such as face recognition and object tracking. For example, in a face recognition scene, it is necessary to detect the face before extracting features in a specific area to verify identity; similarly, target tracking also needs to detect the target position before performing feature similarity matching to track object. At present, target detection has received a lot of attention from academia and industry, and is widely used in public security, smart cities, and autonomous driving. [0003] The current target detection...

AI Technical Summary

Problems solved by technology

This is undoubtedly not an end-to-end method. The intermediate calculation process of the disparity map will make the application process cumbersome and affect the efficiency of practical applications.

Images