Application of soft Histogram of Oriented Gradient on traffic sign detection
Peyman Hosseinzadeh Kassani, Junhyuk Hyun, Euntai Kim · 2016
The proposal of this study is introducing a new feature, called soft Histogram of Oriented Gradients (SHOG). This feature is designed for traffic sign detection. SHOG differs from traditional (hard) HOG in terms of symmetry information and cell of histogram positions. Unlike hard HOG, SHOG changes the positions of cells to a randomized selection of cells following by symmetry shapes of the traffic sign images. SHOG is implemented on the famous German traffic sign detection benchmark (GTSDB) dataset. Comparing to the conventional HOG feature experimented on GTSDB, SHOG could show better performance while uses smaller feature size.