Search

Semantic segmentation, which refers to pixel-wise classification of an image, is a fundamental topic in computer vision owing to its growing importance in the robot vision and autonomous driving sectors. It provides rich information about objects in the scene such as object boundary, category, and location. Recent methods for semantic segmentation often employ an encoder-decoder structure using deep convolutional neural networks. The encoder part extracts features of the image using several filters and pooling operations, whereas the decoder part gradually recovers the low-resolution feature maps of the encoder into a full input resolution feature map for pixel-wise prediction. However, the encoder-decoder variants for semantic segmentation suffer from severe spatial information loss, caused by pooling operations or stepwise convolutions, and does not consider the context in the scene. In this paper, we propose a novel dense upsampling convolution method based on a guided filter to effectively preserve the spatial information of the image in the network. We further propose a novel local context convolution method that not only covers larger-scale objects in the scene but covers them densely for precise object boundary delineation. Theoretical analyses and experimental results on several benchmark datasets verify the effectiveness of our method. Qualitatively, our approach delineates object boundaries at a level of accuracy that is beyond the current excellent methods. Quantitatively, we report a new record of 82.86% and 81.62% of pixel accuracy on ADE20K and Pascal-Context benchmark datasets, respectively. In comparison with the state-of-the-art methods, the proposed method offers promising improvements.

Point cloud registration is a key problem in computer vision applications and involves finding a rigid transform from a point cloud into another such that they align together. The iterative closest point (ICP) method is a simple and effective solution that converges to a local optimum. However, despite the fact that point cloud registration or alignment is addressed in learning-based methods, such as PointNetLK, they do not offer good generalizability for point clouds. In this stud, we proposed a learning-based approach that addressed existing problems, such as finding local optima for ICP and achieving minimum generalizability. The proposed model consisted of three main parts: an encoding network, an auxiliary module that weighed the contribution of each input point cloud, and feature alignment to achieve the final transform. The proposed architecture offered greater generalization among the categories. Experiments were performed on ModelNet40 with different configurations and the results indicated that the proposed approach significantly outperformed the state-of-the-art point cloud alignment methods.

Make and model recognition (MMR) of vehicles plays an important role in automatic vision-based systems. This paper proposes a novel deep learning approach for MMR using the SqueezeNet architecture. The frontal views of vehicle images are first extracted and fed into a deep network for training and testing. The SqueezeNet architecture with bypass connections between the Fire modules, a variant of the vanilla SqueezeNet, is employed for this study, which makes our MMR system more efficient. The experimental results on our collected large-scale vehicle datasets indicate that the proposed model achieves 96.3% recognition rate at the rank-1 level with an economical time slice of 108.8 ms. For inference tasks, the deployed deep model requires less than 5 MB of space and thus has a great viability in real-time applications.

Human computer interaction is a growing field in terms of helping people in their daily life to improve their living. Especially, people with some disability may need an interface which is more appropriate and compatible with their needs. Our research is focused on similar kinds of problems, such as students with some mental disorder or mood disruption problems. To improve their learning process, an intelligent emotion recognition system is essential which has an ability to recognize the current emotional state of the brain. Nowadays, in special schools, instructors are commonly use some conventional methods for managing special students for educational purposes. In this paper, we proposed a novel computer aided method for instructors at special schools where they can teach special students with the support of our system using wearable technologies.

Vehicle analysis involves license-plate recognition (LPR), vehicle-type classification (VTC), and vehicle make and model recognition (MMR). Among these tasks, MMR plays an important complementary role in respect to LPR. In this paper, we propose a novel framework for MMR using local tiled deep networks. The frontal views of vehicle images are first extracted and fed into the local tiled deep networks for training and testing. A local tiled convolutional neural network (LTCNN) is proposed to alter the weight sharing scheme of CNN with local tiled structure. The LTCNN unties the weights of adjacent units and then ties the units k steps from each other within a local map. This architecture provides the translational, rotational, and scale invariance as well as locality. In addition, to further deal with the colour and illumination variation, we applied the histogram oriented gradient (HOG) to the frontal view of images prior to the LTCNN. The experimental results show that our LTCNN framework achieved a 98% accuracy rate in terms of vehicle MMR.