Object Detection Dataset : Navigation Assistance for the Visually Impaired People using YOLOv11

The Dataset contains annotated images from different aspects, considering a visually impaired person’s perspective. The original images were sliced from the recorded video sequence, which were recorded in various locations, including pedestrian pathways and roadside walkways. The different types of viewing perspectives, scene variability and overly crowd density have also been considered during the recording.

The images were annotated in Roboflow using Instance Segmentation. There are 22 object classes such as Bin, Building Pillar, Bus, CNG (Three-wheeler), Car, Cycle, Electric Pole, Food Van, Food cart, Footpath, Leguna, Motorcycle, Obstacle, Parking Cone, Person, Pickup, Rickshaw, Stairs, Tree, Truck, Van, and Van gari (manually driven). The objects which are visible in the range of 20m-30m were annotated.

Dataset details:

Training Set: 86%, 7023 images
Validation Set: 7%, 551 images
Testing Set: 7%, 540 images

Preprocessing :
Auto-Orient: Applied
Resize: Stretch to 640x640

Augmentations:
Flip: Horizontal
Rotation: Between -10° and +10°
Brightness: Between -20% and +20%
Exposure: Between -10% and +10%
Blur: Up to 0.8px
Noise: Up to 0.1% of pixels