Existing learning-based algorithms have a certain potential in visual odometry. In this work, we propose the solution of the learning-based method, which contains the attention mechanism and pose graph optimization. We set a self-supervised network as our backbone to cope with image data and error-heavy estimation pose for pose correction. The pre-processing camera poses involved in the network can provide prior information. Combining the advantages of the abundant feature information and efficient attention mechanism, we design a geometric attention module that is sensitive to geometrical structure from images to accurately regress the rotation matrix. Then we improve the loss function with the weights of the attention module to consider the diversity of the data. Experimental results demonstrate the effectiveness and reliability of our approach on the public datasets KITTI with monocular task and stereo task. In comparison, the proposed method is superior to the existing methods in the translation component. In the self-supervised network, learning an attention mechanism can extract an effective connect relation of feature maps. We conduct ablation experiments under the self-supervised network backbone setting different strategies, and conclude that the proposed attention module is applicable to various sequences, and provide loss function improvements on the visual odometry task.

Tracking and mapping functions in a monocular SLAM system remain active due to their challenging nature. In this paper, we propose a novel approach to perform the accurate and robust ego-motion estimation and provide the detail-preserving reconstruction in indoor environments. More specifically, we design a new algorithm called synchronous event measurement (SEM) to create event-based difference images (EDIs) so as to highlight frame-to-frame (F2F) difference. The observation indicates that F2F difference is highly correlated with the camera's motion change. We hereby feed EDIs into a deep convolutional neural network, in order to infer ego-motion of the camera. Subsequently, based on a monocular reconstruction framework (REMODE), we devise an algorithm named event region search or briefly ERS, to reduce possibility of mismatch on the depth estimation stage. Evaluations on a variety of datasets demonstrate the satisfactory performance of our proposed method: the ego-motion estimation is more accurate than some geometric based Visual Odometry (VO) and learning based approaches. The results are robust under extreme situations, such as brightness variation and motion blur. Meanwhile, our approach can provide more precise depth map with relatively rich textural information.

3-D pose estimation for texture-less objects remains a challenging problem. Previous works either focus on a template matching method to find the nearest template as a candidate, or construct a Hough forest, which utilizes the offset of patches to vote for the object location and pose. By contrast, in this paper, we propose a comprehensive framework to directly regress 3-D poses for the candidates, in which a convolutional neural network-based triplet network is trained to extract discriminating features from the binary images. To make the features suitable for the regression task, a pose-guided method and a regression constraint are employed with the constructed triplet network. We show that the constraint reaches the goal of creating the correlation between the features and 3-D poses. Once the expected features are obtained, the object pose could be efficiently regressed, by training a regression network with a simple structure. For symmetric objects, depth images are treated as an additional channel to feed the triplet network. Experiments on the LineMOD and our own datasets demonstrate our method with high regression precision and efficiency.