Object Detection Algorithms in Computer Vision

Object detection in computer vision encompasses the automatic identification and localisation of objects within images or video streams. Early approaches relied on handcrafted features and shallow classifiers, but the advent of convolutional neural networks (CNNs) ushered in a wave of powerful deep-learning techniques. Two main design paradigms have emerged: two-stage detectors that generate region proposals before classification, and one-stage detectors that predict object classes and bounding boxes in a single pass. Modern frameworks balance speed and accuracy, leveraging backbone architectures for feature extraction, multi-scale feature pyramids for handling objects of varying sizes, and refined loss functions to improve localisation. Evaluation typically employs metrics such as mean average precision and intersection over union to gauge detection quality. Recent advances address challenges including small-object detection, occlusion, real-time inference on edge devices and adaptation to diverse application domains such as autonomous vehicles, remote sensing and urban safety systems.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Object Detection Algorithms in Computer Vision publication trend

The graph below shows the total number of articles in object detection algorithms in computer vision across all publications each year (not limited to Nature Index journals).

Technical terms

Object detection: The task of locating and classifying instances of predefined object categories in an image or video.

Convolutional neural network (CNN): A deep-learning model employing convolutional layers to extract hierarchical feature representations from visual data.

Bounding box: A rectangular annotation specifying the position and size of a detected object within an image.

Two-stage detector: A framework that first generates region proposals and then classifies and refines each proposal in a separate stage.

One-stage detector: A unified architecture that predicts object classes and bounding boxes simultaneously across dense sampling of locations.

Intersection over union (IoU): A metric measuring the overlap between predicted and ground-truth bounding boxes, used in both training and evaluation.

Mean average precision (mAP): A standard performance measure computed as the average of precision values at various recall levels across object categories.

References

1. A Comprehensive Review of YOLO Architectures in Computer Vision: From YOLOv1 to YOLOv8 and YOLO-NAS. Machine Learning and Knowledge Extraction (2023).
2. A Survey of Deep Learning-Based Object Detection. IEEE Access (2019).
3. A Comparative Analysis of Object Detection Metrics with a Companion Open-Source Toolkit. Electronics (2021).
4. DC-YOLOv8: Small-Size Object Detection Algorithm Based on Camera Sensor. Electronics (2023).