Target detection method and device and corresponding model training method and device

A target detection and model technology, applied in the field of target detection, can solve the problems of not being able to identify large-category targets, and not being able to identify large or medium-sized vehicles, so as to improve detection accuracy, realize large-category target detection, and alleviate long tails effect of effect

CN111241969APending Publication Date: 2020-06-05BEIJING SANKUAI ONLINE TECH CO LTD
0 Cites 6 Cited by

Patent Information

Authority / Receiving Office
CN · China
Current Assignee / Owner
Publication Date
2020-06-05

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention discloses a target detection method and device and a corresponding model training method and device. The target detection model training method comprises the following steps: analyzing original training data with labels, and determining label category distribution contained in the original training data; performing data augmentation on the original training data according to the annotation category distribution to obtain training data; and performing iterative training on a target detection model according to the training data until a training stop condition is met. The method has the beneficial effects that data augmentation can be carried out on few sample categories in the original training data according to the distribution condition of the original training data categories; and iterative training is carried out according to the augmented original data, so that the long-tail effect of a detection result caused by imbalance of the original data is relieved, the detection accuracy of multiple categories is improved, and large-category target detection is realized.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The present application relates to the field of target detection, in particular to a target detection method and device and a corresponding model training method and device. Background technique

[0002] In the existing automatic driving scene, it is a common method to use lidar to detect the target in the scene, determine the relative position between the target and the automatic driving device, the type of the target and other information. But the target category that existing technology can detect is less, for example, can only judge that a target is a car, but can't identify specifically whether it is a large car or a medium-sized car, whether it is a van or an SUV (sport utility vehicle), that is, only Scenes that can support small-category recognition cannot effectively identify large-category targets. Contents of the invention

[0003] In view of the above problems, the present application is proposed to provide an object detection method and...

Examples

Embodiment Construction

[0048]Using lidar to detect targets can use volume element-based neural network voxelnet, pointpillars-based algorithm pointpillars, sparse convolution SECOND (Sparsely Embedded Convolutional Detection, sparsely embedded convolution detection) and other schemes, these schemes are based on volume elements Feature extraction methods such as voxelized features or 3D convolutions are used to extract features from point clouds, and then use RPN (RegionProposalNetwork, feature stacking to perform region generation network) from the perspective of bev (bird's eye view) to perform target regression and Classification. What needs to be explained here is that the regression in the field of target detection refers to solving the determination of the detection frame as a mathematical regression problem, and the detection frame refers to the smallest bounding box that can determine the target, for example, in 2D scenes, it is usually A rectangle, usually a cuboid in a 3D scene.

[0049] H...