# Objects as Universal Geolocation Cues: A Computer Vision Approach

## Full Text

This Markdown is an automated text extraction for search and reading tools. The [hosted PDF](ObjectsAsUniversalGeolocationCues.pdf) is the authoritative publication source.

Source PDF: [ObjectsAsUniversalGeolocationCues.pdf](ObjectsAsUniversalGeolocationCues.pdf)
Source SHA-256: `e422d15398c29ff780415685e0719f3f807eef428dba989fc69fa480bde07fd5`
Generated with `pdftotext -nopgbrk`; formatting and reading order may differ from the PDF.

---

Objects as Universal Geolocation Cues:
A Computer Vision Approach
Kanwal Aftab and Mark Scanlon
School of Computer Science, University College Dublin, Ireland.
kanwal.aftab@ucdconnect.ie, mark.scanlon@ucd.ie
MOTIVATION

INSPIRATION

The WHY: Narrow down the search space to assist law enforcement in
combating crimes such as human trafficking and child exploitation.

1) “Trace an Object” is a crowdsourcing campaign by Europol that enlists
the public to identify everyday items such as furniture, logos, or locations
appearing in the background of child sexual abuse material, helping law
enforcement generate new investigative leads [5].

The WHAT: Victims are often photographed in hotel rooms, and
traffickers share these images within criminal networks and online
advertisements. Identifying the hotel from such images can help
determine where a victim was photographed and reveal potential
trafficking hot spots [1].

The HOW: When a large and representative reference dataset is available,
the task is formulated as a Content-based image retrieval (CBIR) problem.
In the absence of such a dataset, geolocation relies on universal visual
cues to infer location directly from the image.

2) In Computer vision many researchers have explored methods like
skyline detection, landmark recognition, and sun azimuth estimation to
determine location in outdoor environment [6][7].

PLUG TO PLACE

Country-specific brands/logos, language cues, and window-visible
landmarks; group same-room images to propagate contextual signals; and
CBIR based fabric patterns can be explored to improve geographic
prediction

Electrical plug sockets are consistent and recognisable indoor markers, as
each country or region uses specific socket types defined by distinct pin
configurations.

FUTURE DIRECTIONS

1. INDOOR GEOLOCATION CUES

BRAND/LOGO

LANGUAGE

FOOD OUTLET

INSIDE OUT

Figure 1: Plug and Socket Types from Type A to Type N [2].

2. SAME ROOM DETECTION

Figure 2: Worldwide plug type distribution map: Color indicate socket type; stripped areas mean
multiple socket types used in a region[3].

3. PATTERN BASED CBIR

EVALUATION AND RESULTS
Stage

Stage 1

Stage 2

Stage 3

Task/Model

Dataset

Socket Detection
Socket Detection
(YOLOv11-Small, KDataset – 2,328
Fold Crossannotated images +
Validation)
4,074 augmented
Socket Type
Socket Type
Classification
Classification
(Best Model:
Dataset – 12 socket
Xception)
type, 3,187 images

Geolocation

Evaluation Dataset:
Hotels-50K
(TraffickCam) 44,630

Metrics

Results

Precision
Recall
mAP@0.5
mAP@0.5:0.95
Accuracy
Precision
Recall
F1-Score
Accuracy @
Confidence
Thresholds (%)
≥ 70%
≥ 80%
≥ 90%

0.8675
0.7990
0.843
0.5771
0.877
0.894
0.884
0.881

91.61
93.73
96.29

Three-stage pipeline: (1) Socket detection, (2) Socket type classification, and (3) Geolocation

For further details refer to our research paper [4]

REFERENCES
[1] S. S. Bhavanasi and A. Stylianou, “Hotel recognition using object ensembles,” in Proc. IEEE Applied Imagery
Pattern Recognition Workshop (AIPR), 2023.
[2] World Standards, “Plugs and sockets.” [Online]. Available:
https://www.worldstandards.eu/electricity/plugs-and-sockets/
[3] Wikipedia, “Mains electricity by country.” [Online]. Available:
https://en.wikipedia.org/wiki/Mains_electricity_by_country#/media/File:World_map_of_electrical_mains_
power_plug_types_used.svg
[4] K. Aftab, G. Adams, and M. Scanlon, “Plug to Place: Indoor multimedia geolocation from electrical sockets
for digital investigation,” arXiv preprint arXiv:2512.16620, 2025.
[5] Europol, “Stop Child Abuse.” [Online]. Available: https://www.europol.europa.eu/stopchildabuse
[6] P. Kakar and N. Sudha, “Authenticating image metadata elements using geolocation information and sun
direction estimation,” in Proc. IEEE Int. Conf. on Multimedia and Expo, 2012.
[7] S. Ramalingam et al., “Skyline2gps: Localization in urban canyons using omni-skylines,” in Proc. IEEE/RSJ
Int. Conf. on Intelligent Robots and Systems, 2010.