Traditional Detection Models

Faster R-CNN Explained for Object Detection Tasks (DigitalOcean)

Normally, the traditional detection pipeline follows three major tasks:

• Region Proposal: identify candidate regions potentially containing objects (> 2.000).

  • Selective Search

  • Edge Boxes

• Feature Extraction: fixed-length feature vector using image descriptors.

• Classification: to assign the background class or one of the object classes to detect. Initially using an SVM classifier.