search

lvyongshjd / A-new-benchmark-dataset-for-sweeping-robot

12 stars 1 forks source link

readme

Object Detection for Sweeping Robots In the Home Scene (ODSR-IHS)：A new benchmak dataset

  1. About ODSR-IHS

Object detection plays an important role in computer vision. It has a variety of applications, including security detection, vehicle recognition, and service robots. With the continuous improvement of public databases and the development of deep learning, object detection has witnessed significant breakthroughs. However, the object detection of sweeping robots during operations should consider various factors, including the camera angle, indoor scenery, and identification of object category. To the best of our knowledge, no corresponding database on these conditions has been developed.

In this study, we propose a large-scale publicly available benchmark dataset called object detection for sweeping robots in home scenes (ODSRIHS). The dataset has 6,000 images and 16,409 instances of 14 object categories.

  1. Collection Method

The data mainly reflect the complex indoor environments, with various target shapes and a considerable proportion of flexible and small targets. In the process of data collection, the following principles are satisfied:

(1) The ratio of sunny to cloudy weather is not higher than 9:1. (2) The ratio of day to night is not higher than 4: 1. (3) The number of indoor scenes is not less than 50. (4) The acquisition angle is consistent with that of an actual sweeping robot

These 14-object classes include bed, sofa, cabinet, stool, table, close stool, trashcan, slippers, wire, socks, carpet, book, feces, and curtain.

  1. Some examples and calibration

The data set is divided into 4 folders, respectively named Annotations, ImageSets, JPEGImages, labels. In the Annotations folder, the xml format tag file corresponding to each picture is saved. The specific information is the file name, image size, object category and bounding box coordinate information of each picture. The JPEGImages folder contains all the picture information, which can be divided into training set, validation set and test set according to specific needs. The visualization of some data sets is shown in the figure.

image-20210507105136152

image-20210507105149514

  1. Learn more

If you use this dataset, please cite these papers:

[1]Lv Y, Fang Y, Chi W, et al. Object detection for sweeping robots in home scenes (ODSR-IHS): a novel benchmark dataset[J]. IEEE Access, 2021, 9: 17820-17828.(https://ieeexplore.ieee.org/document/9333554)

[2]Lv Y, Liu J, Chi W, et al. An inverted residual based lightweight network for object detection in sweeping robots[J]. Applied Intelligence, 2022, 52(11): 12206-12221.

[3]Y. Lv, Y Zhou, W Chi, et al. YOLO_SRv2 An evolved version of YOLO_SR[J] Engineering Applications of Artifical Intelligence(early access), 2023)

(1) In the latest YOLOSRv2 paper, we have expanded the dataset. The number of pictures was increased from 6,000 to 11,104 and the number of instances was increased from 16,409 to 28,381. Due to the large size of the files, we have adopted the sharing method of Baidu Netdisk. The specific link is as follows：https://pan.baidu.com/s/1y8tWlsduJO1yaWllyV-aYQ and extraction code：sif7.

(2) Explains the dataset and how the data is collected

(3) Use mainstream algorithms based on deep learning to verify the data set

(4) Transplant the trained model to the chip for verification