http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Bicycle Detection by Variable Window Approach of 2DHOG Descriptors
Ryo Kawanami,Kousuke Matsushima 대한전자공학회 2017 대한전자공학회 학술대회 Vol.2017 No.1
Recently, various methods are proposed about bicycle detection. However, none of them are universal due to factors such as pose variation, scale changes, and rotation. In order to solve these problems, 2-dimensional HOG (2DHOG) was proposed as a feature extraction method. This method has advantage of high recognition rate, whereas the processing time is much longer. In this paper, we propose an effective method to reduce the processing time with variable window approach in 2DHOG. In variable window approach, pixels are decided randomly in an image. These pixels are used as a center pixel in elliptical region of 2DHOG. Furthermore, the elliptical region size is able to do variable. We did an experiment about bicycle detection in three directions (vertical, diagonal, and horizontal). The experimental result shows that proposed method performs better than the other methods.
Advanced Rotation-Invariant Feature Detection Method for Pedestrian Recognition
Toshiki Yahiro,Kousuke Matsushima 대한전자공학회 2017 대한전자공학회 학술대회 Vol.2017 No.1
In the technology of Advanced Safety Vehicle (ASV), pedestrian detection is an important element. There are techniques using image features and classifiers, but pedestrian regions often include various poses such as rotation. That’s why we cannot always get similar feature values for the same features. The rotation-invariant Histogram of Oriented Gradients(RI-HOG) is a candidate to solve this problem, but this method is not highly accurate because it is not optimized for pedestrian detection. Accordingly, we improved calculating method of the RI-HOG for pedestrian detection and compared this proposed method with the conventional method. As a result, recognition rate was turned from 57.50% to 89.56%.