Method and apparatus for model training and object perception

CN119313893APending Publication Date: 2025-01-14RUILIAN XINGCHEN (BEIJING) TECH CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202310815181.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-07-04
Publication Date
2025-01-14

AI Technical Summary

Technical Problem

In the prior art, it is difficult to accurately detect the occlusion properties of the occluded object in object perception, resulting in large distance measurement errors and affecting perception performance.

Method used

By determining the projection bounding box of the candidate object based on the 3D point cloud data and matching it with the 2D bounding box in the sample image, the occlusion properties of the sample object are determined, and then the perception model is trained to detect the 2D bounding box and 2D bounding box of the object in the input image. Occlusion properties.

Benefits of technology

It achieves accurate detection of the occlusion attributes of occluded objects, improves the accuracy and efficiency of object perception, reduces the cost of data annotation and improves the accuracy of annotation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119313893A_ABST
    Figure CN119313893A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides a method, a device, equipment and a medium for model training and object perception. The method comprises the steps of determining a plurality of projection bounding boxes of a plurality of candidate objects in a two-dimensional projection image based on size marking information of the plurality of candidate objects in three-dimensional point cloud data; determining a sample projection bounding box matched with the sample object from a plurality of projection bounding boxes of the plurality of candidate objects, wherein the sample object is marked with a two-dimensional bounding box in the sample image; determining a shielding attribute of the sample object by comparing the sample projection bounding box with the two-dimensional bounding box; and training a perception model based on the sample image marked with the two-dimensional bounding box and the occlusion attribute, the perception model being configured to detect the two-dimensional bounding box of the object in the input image and the occlusion attribute of the object. Therefore, the perception model capable of accurately detecting the occlusion attribute of the occluded object can be obtained. Based on the perception model, the bounding box of the shielded object can be restored.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Cited By

  • Position determination method and device

    CN120147428A