Lifelong mapping presents unique challenges to household robots which operate in the same environment over long durations. One is the growth of redundant information in the map as it evolves over time, which can easily overwhelm the limited computation resources of a household robot. Another is the possibility of mapping errors. An error in robot pose estimate, which if not corrected fast enough, will result in incorrect occupancy and semantic representation, rendering the map unusable. Finally, for a lifelong mapping system where the map is updated continuously, avoiding these errors altogether is infeasible. In this paper, we present a comprehensive overview of novel strategies for eliminating redundant information from the map and preventing and correcting mapping errors. We also present a detailed evaluation of these novel strategies on 10,000 robots running in indoor environments across different geographic locations of the world to demonstrate map stability and accuracy over time.

In this paper we introduce a novel SLAM system for 2-D indoor environments that relies only on limited sensing. Our fully autonomous system uses only the trajectory of the robot around walls and objects in the environment as landmarks and is capable of robust and long-term exploration and mapping of a broad range of household floor plans. Rank-deficient and full-rank factors are created when the robot observes existing trajectory-based landmarks, and they are filtered and added in a pose graph, which is optimized periodically. The mission space is mapped by efficient adaptive local mapping algorithms. The proposed SLAM system has been extensively tested in various scenarios, and experimental results show its robustness and accuracy.

ImageNet-pretrained networks have been widely used in transfer learning for monocular depth estimation. These pretrained networks are trained with classification losses for which only semantic information is exploited while spatial information is ignored. However, both semantic and spatial information is important for per-pixel depth estimation. In this paper, we design a novel self-supervised geometric pretraining task that is tailored for monocular depth estimation using uncalibrated videos. The designed task decouples the structure information from input videos by a simple yet effective conditional autoencoder-decoder structure. Using almost unlimited videos from the internet, networks are pretrained to capture a variety of structures of the scene and can be easily transferred to depth estimation tasks using calibrated images. Extensive experiments are used to demonstrate that the proposed geometric-pretrained networks perform better than ImageNet-pretrained networks in terms of accuracy, few-shot learning and generalization ability. Using existing learning methods, geometric-transferred networks achieve new state-of-the-art results by a large margin. The pretrained networks will be open source soon(1).

Initialization is essential to monocular Simultaneous Localization and Mapping (SLAM) problems. This paper focuses on a novel initialization method for monocular SLAM based on planar features. The algorithm starts by homography estimation in a sliding window. It then proceeds to a global plane optimization (GPO) to obtain camera poses and the plane normal. 3D points can be recovered using planar constraints without triangulation. The proposed method fully exploits the plane information from multiple frames and avoids the ambiguities in homography decomposition. We validate our algorithm on the collected chessboard dataset against baseline implementations and present extensive analysis. Experimental results show that our method outperforms the fine-tuned baselines in both accuracy and real-time.

This paper presents a method of head pose estimation based on Active Shape Model (ASM) and stereo vision. The relationship between image coordinate system and global coordinate system is outlined in order to get prepared for mapping key landmarks from the former one to the latter one. We utilize Active Shape Model to locate 77 facial landmarks in both left and right images, and implement stereo matching by obtaining disparities directly from the landmarks in the same polar lines. The advantages of the algorithm in accuracy and real-time are manifested by experiments. The head pose is decomposed into pith, yaw and roll, and we use geometric relationship between facial landmarks and the angle around each axis of global coordinate system to estimate head pose. In addition, roll angle compensation method is brought up to expand roll detection range. Finally, we verify the rationality of our proposed head pose estimation method by experiments.

This paper presents a method to recognise traffic police and their command actions based on Kinect. To start with, the way to detect command actions of traffic police is stated in detail. Next, the methodology for judging whether the man conducting command actions is a traffic policeman is presented. At last, the experimental results are shown to verify the rationality and validity of the proposed algorithms in different circumstances. The method of the relative relationship between human body skeletons is applied to classify different actions and it is favorable for solving the problem resulted from relative movements between Kinect and the traffic police. In addition, we employ the technique of analyzing the colors of the police uniform in the different parts of the traffic police by means of combining 3-D space and 2-D space of Kinect to estimate whether the man is a true police. Also, we convert RGB color space to Lab color space to do color segmentation so that the effects of recognition are difficult to be influenced by illumination. Finally, the results of different kinds of experiments are discussed in this paper.