Target object classification method and device

Abstract

The invention discloses a target object classification method and device, and the method comprises the steps: determining a second region of each target object in a to-be-classified image and a third region of each target object in a to-be-classified depth image through obtaining a first region corresponding to each target object in to-be-classified point cloud data, the to-be-classified image and the to-be-classified depth image; and for each target object, inputting a sub-image of the target object in a second area in the to-be-classified image and a sub-image of the target object in a third area in the to-be-classified depth image into the trained classification model, and determining a classification result of each target object. According to the method, the weights of the classification results of the point cloud bounding box and the image bounding box of the target object do not need to be preset respectively, the situation that the classification results are wrong due to the fact that the weights of all the bounding boxes are preset is avoided, and the accuracy of the classification results of the target object is improved.

Description

technical field [0001] This specification relates to the field of unmanned driving technology, in particular to a method and device for object classification. Background technique [0002] At present, during the driving process of the unmanned driving equipment, the objects in the surrounding environment are usually classified, so as to determine the motion strategy of the unmanned equipment based on the classification results of each object to ensure the driving safety of the unmanned equipment. Wherein, the target object can usually be an object such as an obstacle, a signboard, a signal light, etc. that will affect the driving of the unmanned driving device. [0003] In the prior art, object classification can generally be performed based on point cloud data and image data. Specifically, firstly, object detection is performed on the acquired point cloud data and image data respectively, and point cloud bounding boxes of each target object in the point cloud data and imag...

AI Technical Summary

Problems solved by technology

[0004] Since the classification results are more dependent on the image texture information, the weight of the recognition results of the preset image bounding boxes is often higher. Therefore, when the target detection of the image data is missed, the determined The classification result of the target object is not accurate enough, which affects the driving safety of unmanned equipment

Images