Visual SLAM (VSLAM) has been developing rapidly due to its advantages of low-cost sensors, the easy fusion of other sensors, and richer environmental information. Traditional visionbased SLAM research has made many achievements, but it may fail to achieve wished results in challenging environments. Deep learning has promoted the development of computer vision, and the combination of deep learning and SLAM has attracted more and more attention. Semantic information, as high-level environmental information, can enable robots to better understand the surrounding environment. This paper introduces the development of VSLAM technology from two aspects: traditional VSLAM and semantic VSLAM combined with deep learning. For traditional VSLAM, we summarize the advantages and disadvantages of indirect and direct methods in detail and give some classical VSLAM open-source algorithms. In addition, we focus on the development of semantic VSLAM based on deep learning. Starting with typical neural networks CNN and RNN, we summarize the improvement of neural networks for the VSLAM system in detail. Later, we focus on the help of target detection and semantic segmentation for VSLAM semantic information introduction. We believe that the development of the future intelligent era cannot be without the help of semantic technology. Introducing deep learning into the VSLAM system to provide semantic information can help robots better perceive the surrounding environment and provide people with higher-level help.

After decades of development, LIDAR and visual SLAM technology has relatively matured and been widely used in the military and civil fields. SLAM technology enables the mobile robot to have the abilities of autonomous positioning and mapping, which allows the robot to move in indoor and outdoor scenes where GPS signals are scarce. However, SLAM technology relying only on a single sensor has its limitations. For example, LIDAR SLAM is not suitable for scenes with highly dynamic or sparse features, and visual SLAM has poor robustness in low-texture or dark scenes. However, through the fusion of the two technologies, they have great potential to learn from each other. Therefore, this paper predicts that SLAM technology combining LIDAR and visual sensors, as well as various other sensors, will be the mainstream direction in the future. This paper reviews the development history of SLAM technology, deeply analyzes the hardware information of LIDAR and cameras, and presents some classical open source algorithms and datasets. According to the algorithm adopted by the fusion sensor, the traditional multi-sensor fusion methods based on uncertainty, features, and novel deep learning are introduced in detail. The excellent performance of the multi-sensor fusion method in complex scenes is summarized, and the future development of multi-sensor fusion method is prospected.

In the face of large-scale environmental mapping requirements, through the use of lightweight and inexpensive robot groups to perceive the environment, the multi-robot cooperative (V)SLAM scheme can resolve the individual cost, global error accumulation, computational load, and risk concentration problems faced by single-robot SLAM schemes. Such schemes are robust and stable, form a current research hotspot, and relevant algorithms are being updated rapidly. In order to enable the reader to understand the development of this field rapidly and fully, this paper provides a comprehensive review. First, the development history of multi-robot collaborative SLAM is reviewed. Second, the fusion algorithms and architectures are detailed. Third, from the perspective of machine learning classification, the existing algorithms in this field are discussed, including the latest updates. All of this will make it easier for readers to discover problems that need to be studied further. Finally, future research prospects are listed.

Because of the broad application of human action recognition technology, action recognition has always been a hot spot in computer vision research. The Long Short-Term Memory (LSTM) network is a classic action recognition algorithm, and many effective hybrid algorithms have been proposed based on basic LSTM infrastructure. Although some progress has been made in accuracy, most of those hybrid algorithms have to have more and more complex structures and deeper network levels. After analyzing the structure of the classic LSTM from the perspective of control theory, we determined that the classic LSTM could strengthen the differential characteristics of human action recognition technology to reflect the change of speed. Thus, an improved LSTM structure with an input differential characteristic module is proposed. Furthermore, in this article, we considered the influence of first-order and second-order differential on the extraction of movement pose information, that is, the influence of movement speed and acceleration on action recognition. We designed four different LSTM units with first-order and second-order differential. Moreover, the experiments were performed for the four units on three common datasets repeatedly. We found that the LSTM network with the input differential feature module proposed in this article can effectively improve action recognition accuracy and stability without deepening the complexity of the network and can be used as a new basic LSTM network architecture.

A static environment is a prerequisite for the stable operation of most visual SLAM systems, which limits the practical use of most existing systems. The robustness and accuracy of visual SLAM systems in dynamic environments still face many complex challenges. Only relying on semantic information or geometric methods cannot filter out dynamic feature points well. Considering the problem of dynamic objects easily interfering with the localization accuracy of SLAM systems, this paper proposes a new monocular SLAM algorithm for use in dynamic environments. This improved algorithm combines semantic information and geometric methods to filter out dynamic feature points. Firstly, an adjusted Mask R-CNN removes prior highly dynamic objects. The remaining feature-point pairs are matched via the optical-flow method and a fundamental matrix is calculated using those matched feature-point pairs. Then, the environment’s actual dynamic feature points are filtered out using the polar geometric constraint. The improved system can effectively filter out the feature points of dynamic targets. Finally, our experimental results on the TUM RGB-D and Bonn RGB-D Dynamic datasets showed that the proposed method could improve the pose estimation accuracy of a SLAM system in a dynamic environment, especially in the case of high indoor dynamics. The performance effect was better than that of the existing ORB-SLAM2. It also had a higher running speed than DynaSLAM, which is a similar dynamic visual SLAM algorithm.