Improvement of Pedestrian Detection Algorithm Based on YOLO
Download as PDF
DOI: 10.23977/meimie.2019.43009
Author(s)
Xuan Li, Jing Li
Corresponding Author
Jing Li
ABSTRACT
The non-maximum suppression algorithm is usually used in the post-position of deep learning object detection algorithm. It suppresses the detection boxes with high overlap rate while using the algorithm. In order to avoid the missed and false detection caused by non-maximum suppression algorithm, an improved maximum value suppression algorithm is proposed. When the IOU of the suppression window and the suppressed window is greater than the given threshold, the confidence multiply by the penalty factor instead of discarding it directly. After multiple iterations, we need to remove the lower scores detection boxes. Experiments show that the YOLO V2 deep learning model with improved algorithm has improved accuracy on different data sets as well as strong versatility and robustness.
KEYWORDS
Deep learning, object detection, detection boxes, penalty factor, YOLO V2