Due to the rise of convolutional neural network, pedestrian detection has achieved great success. However, the features of many existing methods are not fully utilized, which results in unsatisfactory detection results. A new pedestrian detection model which is named of SE-Faster R-CNN is proposed in this paper. By adding SENet block to Faster RCNN, it can strengthen the expressive force of feature. Then, the GN-Faster R-CNN, which is generated by adding the normalization layer -- Group Normalization layer to Faster R-CNN, is proposed. The proposed architecture is trained and tested on Caltech dataset. In addition, VGG16 model and ZF model are used as the backbone structure of detection network. A comparative experiment is implemented to compare the effectiveness of the two optimization methods. It can be seen from the experimental results that, after adding SENet, the miss rates of ZF model and VGG16 model were reduced by 0.392% and 0.999%, respectively. After adding the GN layer, the miss rate of VGG16 model was reduced by 0.665%, while the miss rate of ZF model was increased by 2.093%.
Due to the rise of deep learning, person re-identification has become a research hotspot in computer vision. For most person re-identification algorithm, softmax function is used as loss function which could increase the distance of interclasses, but has a bad convergence performance for the distance of intra-classes. Therefore, a person re-identification model based on multi-loss optimization is proposed by adding center loss. Center loss has the function of reducing intraclass distance, which makes up for the shortcoming of softmax loss. Two models are selected for comparative experiment to prove the effectiveness of our method. One is the re-ranking person re-identification model with kreciprocal coding, which is named IDE_ResNet-50+Jaccard. The other is the person re-identification model without kreciprocal coding, which is named IDE_ResNet-50. The experiments perform on the Market-1501 dataset, and the result shows that our method has a better result than the original model, which gains an increase of 1.25% and 0.63% in mAP and rank-1 accuracy for IDE_ResNet-50+Jaccard model. For the IDE_ResNet-50 model, the accuracy of mAP and rank1 increased by 1.86% and 0.18%, respectively.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.