RGB-D data-based 3D target detection method and device, equipment and storage medium

Abstract

The invention discloses a 3D target detection method, device and equipment based on RGB-D data and a computer storage medium, and the method comprises the steps: obtaining the RGB-D data and an RGB image of a target container, and converting the RGB-D data into 3D point cloud data; inputting the RGB image into a neural network to obtain a two-dimensional bounding box of a target container; inputting the two-dimensional bounding box of the target container and the 3D point cloud data into a 3D instance segmentation network to obtain point cloud coordinates of the target container; and inputtingthe point cloud coordinates into a 3D bounding box estimation network to extract the spatial position and attitude information of the target container so as to realize fusion perception and identification of the target container. Real-time 3D target detection and positioning of a target container can be realized.

Description

technical field [0001] The present invention relates to the field of computer information technology, in particular to a 3D object detection method, device, equipment and storage medium based on RGB-D data. Background technique [0002] Due to the characteristics of fast speed, low cost and easy management of container transportation, modern container logistics has been continuously developed. With the continuous development of computer control technology and machine vision technology, container tire cranes are also developing in the direction of mechatronics, intelligence, energy saving and high efficiency. At present, many methods for container intelligent loading and unloading have emerged, such as 3D object detection methods. [0003] However, the inventor found in the process of implementing the present invention that the method for intelligent loading and unloading of containers has the following defects: 1. Although the cargo can be identified, it cannot accurately l...

AI Technical Summary

Problems solved by technology

[0003] However, the inventor found in the process of implementing the present invention that the method for container intelligent loading and unloading has the following defects: 1. Although the cargo can be identified, it cannot accurately locate the target cargo in real time

2. It only recognizes the objects in two-dimensional space and marks the frame. The depth information provided has errors, and the obtained frame information is also very insufficient in accuracy, and the accuracy is far from enough

Images