An intelligent car plate detection method can make the travel more convenient and efficient. However, traditional methods are reasonably effective under the specific circumstances or strong assumptions only, and there...
An intelligent car plate detection method can make the travel more convenient and efficient. However, traditional methods are reasonably effective under the specific circumstances or strong assumptions only, and there are few databases for car plate detection. Therefore, a novel real-time car plate detection method based on improved Yolov3 has been proposed. In order to select the more precise number of candidate anchor boxed and aspect ratio dimensions, the K-Means algorithm is utilized. To solve the short of the available car plate database, a car plate database which has 6668 pictures has been established. As shown in the experimental results, the method which is proposed by this paper is better than original Yolov3. Thanks to the car plate database, the proposed method obtained better results even in the situation of inclination, too bright or too dark, different weather and so on.
The parameter optimization of the Object tracking algorithms has achieved excellent performance based on parameter regularization. Adaptive spatial regularization correlation filter can obtain the optimal weight coeff...
The parameter optimization of the Object tracking algorithms has achieved excellent performance based on parameter regularization. Adaptive spatial regularization correlation filter can obtain the optimal weight coefficient for a specific target. Inspired by this, we propose a filter combined temporal regularity with spatial L1 regularization (LTASRCF). The time regularizer is used to remember the data of the previous frame to update the current target position. It performs well in the process of occlusion, deformation and motion blur. Meanwhile, L1 regularization of spatial weights reduces the computational complexity of the coefficients, and improves the robustness of the spatial weight matrix. We have conducted comparative experiments on the OTB-2015 dataset, and the proposed LTASRCF model ranks first among the five popular algorithms, which outperforms by 9.2% and 18.9% than KCF, respectively, and reaches real-time performance.
Deep learning has shown great advantages in biomedical image segmentation. The classic model U-Net uses a stacked encoding-decoding structure of convolution operations for feature extraction and pixel-level classifica...
Deep learning has shown great advantages in biomedical image segmentation. The classic model U-Net uses a stacked encoding-decoding structure of convolution operations for feature extraction and pixel-level classification. The stacking of convolutional layers can expand the receptive field, but it is still a local operation and cannot capture long-distance dependence. Therefore, in this work, we propose a Global Attention Mechanism that combines channel attention module and spatial attention module and integrates different convolutions in it. Besides, we design a residual module for the traditional up and down sampling blocks. And finally, we combine them with U-Net to propose a new global attention network GAU-Net. We perform experiments on the dataset BraTS2018. Our model has increased the mIoU from 0.65 to 0.75 with only 5.4% of U-Net parameters. At the same time, the inference time is also significantly shortened with relatively good performance.
In view of the quick development of modern computer technology, more pictures of natural scenes have been collected and analyzed for application in military fields, scientific research, and computer vision fields rela...
In view of the quick development of modern computer technology, more pictures of natural scenes have been collected and analyzed for application in military fields, scientific research, and computer vision fields related to object detection. However, the sharpness of these pictures collected will be affected by rain and snow, which will reduce the visibility of things in the pictures. This will not only affect the viewing, but also affect scientific research fields such as autonomous driving and video surveillance. Therefore, the research on image rain removing is very practical. An improved algorithm model named pixel-GAN is proposed in this article, which based on conditional Generative Adversarial Networks (cGAN). This new method regards the problem of removing rain from specific images as the style conversion problem. What we improved is that a VGG-16 module is added to the original network to extract perceptual loss as a new part of the loss function. We show the excellent rain-removing ability of the improved algorithm and also illustrate the reason why we add the perceptual loss. Compared with other existing methods, there is no need to set a prior model, the training speed is faster and the proposed model finally generates higher quality pictures.
暂无评论