TY - GEN
T1 - Pedestrian Detection Based on Improved Mask R-CNN Algorithm
AU - Yu, Wenjun
AU - Kim, Sumi
AU - Chen, Fei
AU - Choi, Jaeho
N1 - Publisher Copyright:
© 2021, The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2021
Y1 - 2021
N2 - As the modern society evolves around digital media, object recognition becomes one of the important areas for computer vision. Pedestrian detection particularly draws much attention because it is closely related to everyday life. Recently, pedestrian detection has achieved great success in intelligent monitoring, intelligent driving, and environmental protection. Although, there are several pedestrian detection algorithms based on deep learning, the pedestrian detection is still a huge challenge. Background occlusion, pedestrians’ various changing postures and objects’ occlusion give significant impact on the recognition results; it still brings up much attention. In this paper, to reduce the influence of external factors, we propose a new method based on Mask R-CNN algorithm. The proposed system was evaluated on the Daimler pedestrian dataset for training and on the public Caltech and INRIA pedestrian datasets for testing. The experimental results have showed that the proposed algorithm achieves better detection accuracy than the conventional ones.
AB - As the modern society evolves around digital media, object recognition becomes one of the important areas for computer vision. Pedestrian detection particularly draws much attention because it is closely related to everyday life. Recently, pedestrian detection has achieved great success in intelligent monitoring, intelligent driving, and environmental protection. Although, there are several pedestrian detection algorithms based on deep learning, the pedestrian detection is still a huge challenge. Background occlusion, pedestrians’ various changing postures and objects’ occlusion give significant impact on the recognition results; it still brings up much attention. In this paper, to reduce the influence of external factors, we propose a new method based on Mask R-CNN algorithm. The proposed system was evaluated on the Daimler pedestrian dataset for training and on the public Caltech and INRIA pedestrian datasets for testing. The experimental results have showed that the proposed algorithm achieves better detection accuracy than the conventional ones.
KW - Feature concatenation
KW - Hard negative mining
KW - Mask R-CNN
KW - Pedestrian detection
UR - https://www.scopus.com/pages/publications/85088749941
U2 - 10.1007/978-3-030-51156-2_176
DO - 10.1007/978-3-030-51156-2_176
M3 - Conference paper
AN - SCOPUS:85088749941
SN - 9783030511555
T3 - Advances in Intelligent Systems and Computing
SP - 1515
EP - 1522
BT - Intelligent and Fuzzy Techniques
A2 - Kahraman, Cengiz
A2 - Cevik Onar, Sezi
A2 - Oztaysi, Basar
A2 - Sari, Irem Ucal
A2 - Cebi, Selcuk
A2 - Tolga, A. Cagri
PB - Springer
T2 - International Conference on Intelligent and Fuzzy Systems, INFUS 2020
Y2 - 21 July 2020 through 23 July 2020
ER -