Improving pedestrian detection with selective gradient self-similarity feature
作者：
Highlights：
• We propose the selective gradient self-similarity (SGSS) feature.
• SGSS captures pairwise similarity patterns of local gradient distributions.
• SGSS is a mid-level feature on top of HOG and is complementary to HOG.
• Addition of SGSS gives a significant boost to pedestrian detection accuracy.
• The AdaBoost-based cascade with HOG-SGSS outperforms the linear SVM/HIKSVM.
摘要
Highlights•We propose the selective gradient self-similarity (SGSS) feature.•SGSS captures pairwise similarity patterns of local gradient distributions.•SGSS is a mid-level feature on top of HOG and is complementary to HOG.•Addition of SGSS gives a significant boost to pedestrian detection accuracy.•The AdaBoost-based cascade with HOG-SGSS outperforms the linear SVM/HIKSVM.
论文关键词：Pedestrian detection,Contour description,Self-similarity,Feature selection,Cascade
论文评审过程：Received 21 July 2014, Revised 9 November 2014, Accepted 12 January 2015, Available online 17 January 2015.
论文官网地址：https://doi.org/10.1016/j.patcog.2015.01.005
