Definition
Object detection is a computer vision task that finds each object in an image and gives it a bounding box and a class. A bounding box is a rectangle around the object.
Detector types
Detectors differ in how they find the boxes:
- Two-stage detectors. Faster R-CNN uses a region proposal network that shares image features with the detection network.
- Single-pass detectors. YOLO treats detection as a regression problem. One network predicts the boxes and class probabilities from the full image in one evaluation. The first YOLO model ran at 45 frames per second.
- Transformer detectors. DETR predicts a set of objects with a transformer encoder-decoder. It does not need hand-designed steps such as non-maximum suppression and anchor generation. On COCO, its accuracy and runtime were on par with Faster R-CNN.
The COCO dataset has 328,000 images, 91 object types and 2.5 million labeled instances. Researchers score many detectors on it.
Airfield inspection
Foreign object debris (FOD) is an inanimate object in the movement area. It has no operational or aeronautical function and can be a hazard to aircraft. The EASA aerodrome rules require the operator to have procedures to detect FOD. They do not require a specific technology.
The FAA advisory circular on FOD detection equipment describes electro-optical systems as systems that use video and image processing. Its use is mandatory only for equipment bought with Airport Improvement Program or Passenger Facility Charge funds. It sets test objects, such as a metal cylinder 3.1 cm high and a sphere 4.3 cm in diameter.
A bounding box does not give the area or the length of a defect. The PAVER manual measures alligator cracking by area and long cracks by length.
Frequently Asked Questions
- What does an object detector output?
- An object detector outputs a list of bounding boxes with a class for each box. YOLO, for example, predicts bounding boxes and class probabilities directly from the full image in one evaluation of a single neural network.
- How is an object detector scored?
- The COCO benchmark scores a detector with average precision, averaged over ten IoU thresholds from 0.50 to 0.95 in steps of 0.05. IoU measures how much a predicted box overlaps the labeled box. COCO also reports results for small objects (area under 32 by 32 pixels), medium objects and large objects.
- Does EASA require object detection software for foreign object debris?
- No. Regulation (EU) No 139/2014 requires the aerodrome operator to establish procedures to detect FOD, which include monitoring and inspection of the movement area. The acceptable means of compliance for staff training mention the proper use of detection technologies, if applicable. They do not name object detection software.
- Is a bounding box enough to measure a pavement crack?
- No. A bounding box shows where a defect is, but not its area or length. The PAVER distress manual for asphalt airfields measures alligator cracking in square feet (square meters) and longitudinal and transverse cracks in linear feet (linear meters).
References
- ren-2015-faster-rcnn
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks Ren, He, Girshick, Sun (arXiv:1506.01497)
- redmon-2015-yolo
- You Only Look Once: Unified, Real-Time Object Detection Redmon, Divvala, Girshick, Farhadi (arXiv:1506.02640)
- carion-2020-detr
- End-to-End Object Detection with Transformers Carion et al. (arXiv:2005.12872)
- lin-2014-coco
- Microsoft COCO: Common Objects in Context Lin et al. (arXiv:1405.0312)
- pycocotools-cocoeval
- pycocotools, cocoeval.py (COCO evaluation code) COCO Consortium
- easa-ear-aerodromes
- Easy Access Rules for Aerodromes (Regulation (EU) No 139/2014), including CS-ADR-DSN Issue 7 EASA
- faa-ac-150-5220-24
- AC 150/5220-24, Airport Foreign Object Debris (FOD) Detection Equipment (30 September 2009) FAA
- faa-paver-asphalt
- PAVER Distress Identification Manual for Asphalt Surfaced Airfields FAA
This is a reference entry. TarmacView provides drone inspection services and doesn't sell the equipment or materials described in this glossary.
