Back to results
US 11,776,288 B2Granted
Object detection method, and training method for a target object detection model
AssigneeUbtech Robotics Corp Ltd
InventorsYonghui Cai, Jun Cheng, Jianxin Pang, Youjun Xiong
Life of this patent
Estimating the term…
Abstract
A target object detection model is provided. The target object detection model includes a YOLOv3-Tiny model. Through the target object detection model, low-level information in the YOLOv3-Tiny sub-model can be merged with high-level information therein, so as to fuse the low-level information and the high-level information. Since the low-level information can be further used, the comprehensiveness of target detection is effectively improved, and the detection effect of small targets is improved.
Claims18
A computer-implemented training method for a target object detection model based on a YOLOv3-Tiny model, comprising:
- obtaining a plurality of first training samples, wherein each of the first training samples is an image marked with one or more target objects;
- inputting each of the first training samples to the target object detection model to perform an iterative training to obtain a trained target object detection model;
- and
- storing the trained target object detection model on a non-transitory computer readable storage medium to use the trained target object detection model for an object detection;
- wherein the YOLOv3-Tiny model has an input receiving an image, and the target object detection model comprises:
- a first convolutional layer connected to a first low-level information output end (101) of the YOLOv3-Tiny model to receive and perform convolution on low-level information of the image;
- wherein the first convolutional layer has an output end connected to a first concatenating layer of the YOLOv3-Tiny model, a high-level information output end (201) of the YOLOv3-Tiny model is connected to and outputs high-level information of the image to the first concatenating layer, the first concatenating layer concatenates the low-level information and high-level information and outputs the concatenated information to first high-level information convolutional layers (301) of the YOLOv3-Tiny model, thereby realizing fusing multi-level information;
- and
- wherein the target object detection model further comprises:
- a second convolutional layer connected to a second low-level information output end (102) of the YOLOv3-Tiny model to receive and perform convolution on the low-level information of the image;
- and
- a second concatenating layer connected to the second convolutional layer and another high-level information output end (202) of the YOLOv3-Tiny model, to concatenate information, and wherein convolution is performed on the concatenated information from the second concatenating layer by second high-level information convolutional layers (302) of the YOLOv3-Tiny model.
9 more elements
A non-transitory computer readable storage medium for storing a target object detection model based on a YOLOv3-Tiny model, the YOLOv3-Tiny model having an input receiving an image, the target object detection model comprising:
- a first convolutional layer connected to a first low-level information output end (101) of the YOLOv3-Tiny model to receive and perform convolution on low-level information of the image;
- wherein the first convolutional layer has an output end connected to a first concatenating layer of the YOLOv3-Tiny model, a high-level information output end (201) of the YOLOv3-Tiny model is connected to and outputs high-level information of the image to the first concatenating layer, the first concatenating layer concatenates the low-level information and high-level information and outputs the concatenated information to first high-level information convolutional layers (301) of the YOLOv3-Tiny model, thereby realizing fusing multi-level information;
- and
- wherein the target object detection model further comprises:
- a second convolutional layer connected to a second low-level information output end (102) of the YOLOv3-Tiny model to receive and perform convolution on the low-level information of the image;
- and\
- a second concatenating layer connected to the second convolutional layer and another high-level information output end (202) of the YOLOv3-Tiny model, to concatenate information, and wherein convolution is performed on the concatenated information from the second concatenating layer by second high-level information convolutional layers (302) of the YOLOv3-Tiny model;
- wherein the non-transitory computer readable storage medium further stores a computer program executable on a processor, and the computer program comprises:
- instructions for collecting, by a camera of a service robot, object images;
- instructions for inputting the object images into the target object detection model, and obtaining a coordinate of a target object in the object images through feature extraction and classification;
- and instructions for marking a position of the target object on the object images according to the coordinate.
8 more elements
A computer-implemented object detection method applied to a service robot, comprising:
- collecting, by a camera of the service robot, object images;
- inputting, by an operating system of the service robot, the object images into a pre-built target object detection model based on a YOLOv3-Tiny model, and obtaining a coordinate of a target object in the object images through feature extraction and classification;
- and
- marking, by the operating system, a position of the target object on the object images according to the coordinate;
- wherein the YOLOv3-Tiny model has an input receiving an image, and the target object detection model comprises:
- a first convolutional layer connected to a first low-level information output end (101) of the YOLOv3-Tiny model to receive and perform convolution on low-level information of the image;
- wherein the first convolutional layer has an output end connected to a first concatenating layer of the YOLOv3-Tiny model, a high-level information output end (201) of the YOLOv3-Tiny model is connected to and outputs high-level information of the image to the first concatenating layer, the first concatenating layer concatenates the low-level information and high-level information and outputs the concatenated information to first high-level information convolutional layers (301) of the YOLOv3-Tiny model, thereby realizing fusing multi-level information;
- a second convolutional layer connected to a second low-level information output end (102) of the YOLOv3-Tiny model to receive and perform convolution on the low-level information of the image;
- and
- a second concatenating layer connected to the second convolutional layer and another high-level information output end (202) of the YOLOv3-Tiny model, to concatenate information, and wherein convolution is performed on the concatenated information from the second concatenating layer by second high-level information convolutional layers (302) of the YOLOv3-Tiny model.
7 more elements
Dependencies are inferred from the claim text.