Multi-Environment Object Detection Dataset for Improved Indoor Navigation: Focusing on the Importance of Different Door Handle Types

The object detection dataset consists of 3,292 images captured from various environments using a simple data collection system. Images were taken from different perspectives, including handheld and camera-mounted on an EPW (Environmental Process Workstation), to ensure diversity and varied perspectives. The dataset was divided into 60% for training (1,975 images), 10% for validation (330 images), and 30% for testing (987 images). The image resolution is 512x512x3 pixels, which is lower than a proposed semantic segmentation dataset but still comparable to standard datasets. The dataset contains annotations on the bounding box level and includes eight object categories such as doors, fire extinguishers, key slots, switches, ID readers, and various types of door handles. The dataset addresses the absence of infrequent objects found in other publicly available datasets. Unclassifiable objects are left unannotated.

Folders description:
- images -> dataset images.

Files descriptions:
- AllDataTable -> All the images and the corresponding bounding boxes in a Matlab table format.

- trainingData_datastore.mat -> training datastore in Matlab format. The combined datastore has the images file path and the corresponding bounding boxes.

- testingData_datastore.mat -> testing datastore in Matlab format. The combined datastore has the images file path and the corresponding bounding boxes.

- validationData_datastore.mat -> validation datastore in Matlab format. The combined datastore has the images file path and the corresponding bounding boxes.

Note: files path in the training, validation and testing Matlab files need to be modified to point to the location where the images locate.

Classes:
- Classes ->
'door'
'pull door handle'
'push button'
'moveable door handle'
'push door handle'
'fire extinguisher'
'key slot'
'id_reader'

Paper:
Mohamed, E.; Sirlantzis, K.; Howells, G.; Hoque, S. Optimisation of Deep Learning Small-Object Detectors with Novel Explainable Verification. Sensors 2022, 22, 5596. https://doi.org/10.3390/s22155596