In this paper, a new scene recognition visual descriptor called Enhanced Scale Invariant Feature Transform-based Sparse coding Spatial Pyramid Matching (Enhanced SIFT-ScSPM) descriptor is proposed by combining a Bag of Words (BOW)-based visual descriptor (SIFT-ScSPM) and Gist-based descriptors (Enhanced Gist-Enhanced multichannel Gist (Enhanced mGist)). Indoor scene classification is carried out by multi-class linear and non-linear Support Vector Machine (SVM) classifiers. Feature extraction methodology and critical review of several visual descriptors used for indoor scene recognition in terms of experimental perspectives have been discussed in this paper. An empirical study is conducted on the Massachusetts Institute of Technology (MIT) 67 indoor scene classification data set and assessed the classification accuracy of state-of-the-art visual descriptors and the proposed Enhanced mGist, Speeded Up Robust Features-Spatial Pyramid Matching (SURF-SPM) and Enhanced SIFT-ScSPM visual descriptors. Experimental results show that the proposed Enhanced SIFT-ScSPM visual descriptor performs better with higher classification rate, precision, recall and area under the Receiver Operating Characteristic (ROC) curve values with respect to the state-of-the-art and the proposed Enhanced mGist and SURF-SPM visual descriptors.

Considering that traditional visual navigation cannot be utilised in low illumination and sparse feature environments, a novel visual-inertial integrated navigation method using a Structured Light Visual (SLV) sensor for Micro Aerial Vehicles (MAVs) is proposed in this paper. First, the measurement model based on an SLV sensor is studied and built. Then, using the state model based on error equations of an Inertial Navigation System (INS), the measurement model based on the error of the relative motion measured by INS and SLV is built. Considering that the measurements in this paper are mainly related to the position and attitude information of the present moment, the state error accumulation in traditional visual-inertial navigation can be avoided. An Adaptive Sage-Husa Kalman Filter (ASHKF) based on multiple weighting factors is proposed and designed to make full use of the SLV measurements. The results of the simulation and the experiment based on real flight data indicate that high accuracy position and attitude estimations can be obtained with the help of the algorithm proposed in this paper.

Vision navigation using environmental features has been widely applied when satellite signals are not available. However, the matching performance of traditional environmental features such as keypoints degrades significantly in weakly textured areas, deteriorating navigation performance. Further, the user needs to evaluate and assure feature matching quality. In this paper, a new feature, named Line Segment Intersection Feature (LSIF), is proposed to solve the availability problem in weakly textured regions. Then a combined descriptor involving global structure and local gradient is designed for similarity comparison. To achieve reliable point-to-point matching, a coarse-to-fine matching algorithm is developed, which improves the performance of the point set matching algorithm. Finally, a framework of matching quality evaluation is proposed to assure matching performance. Through the comparison, it is demonstrated that the proposed new feature has superior overall performance especially on correctly matched numbers of keypoints and matching correctness. Also, using real image sets with weak texture, it is shown that the proposed LSIF can achieve improved navigation solutions with high continuity and accuracy.

Autonomous Unmanned Ground Vehicles (UGVs) require a reliable navigation system that works in all environments. However, indoor navigation remains a challenge because the existing satellite-based navigation systems such as the Global Positioning System (GPS) are mostly unavailable indoors. In this paper, a tightly-coupled integrated navigation system that integrates two dimensional (2D) Light Detection and Ranging (LiDAR), Inertial Navigation System (INS), and odometry is introduced. An efficient LiDAR-based line features detection/tracking algorithm is proposed to estimate the relative changes in orientation and displacement of the vehicle. Furthermore, an error model of INS/odometry system is derived. LiDAR-estimated orientation/position changes are fused by an Extended Kalman Filter (EKF) with those predicted by INS/odometry using the developed error model. Errors estimated by EKF are used to correct the position and orientation of the vehicle and to compensate for sensor errors. The proposed system is verified through simulation and real experiment on an UGV equipped with LiDAR, MEMS-based IMU, and encoder. Both simulation and experimental results showed that sensor errors are accurately estimated and the drifts of INS are significantly reduced leading to navigation performance of sub-metre accuracy.

The basic idea of a dead reckoning personal navigator is to integrate incremental motion information in the forms of step length and step direction over time. Considering that the displacement components are estimated for each step-cycle, it is essential to monitor the integrity of these parameters; otherwise, the error accumulation may render the system unstable. In this paper, a two-stage Kalman Filter (KF) augmented by a neural network is developed to facilitate integrity monitoring. The preliminary results, obtained from several tests performed on simulated and real-word data, indicate an 80% success rate in integrity monitoring.

This paper presents further research on the SNUGL pseudolite-based navigation system presented in this journal in 2003. This system has centimetre-level accuracy, but has an error source arising from right-hand circularly polarized (RHCP) transmissions, unlike outdoor Global Positioning System (GPS). The GPS satellites and pseudolites use RHCP signals, and the polarization affects carrier-phase measurements according to the Line-of-Sight (LOS) vectors from transmitters to receivers. The RHCP error is eliminated by a double differencing process in outdoor GPS, but the error remains in the pseudolite-based system because the LOS vectors from transmitters to a reference and user receivers are different for the close transmitter constellations. This paper shows the RHCP effect on the pseudolite-based navigation system through simulations and experiments. It then shows the RHCP-compensation method improves the measurement and position accuracy by over 10%.

Autonomous mobile robot navigation systems are based on three principal kinds of techniques: map-based navigation, map-building-based navigation and mapless navigation. We propose a mapless method for trajectory description in unknown indoor environments. The method uses distance measurements from a 2D laser range finder, digitises the robot's visibility area, eliminates superfluous data and reorients their presentation with laws similar to those used in cellular automata. The landmarks are extracted and organised in a panoramic description called fresco. The frescoes which are validated by means of neighbourhood rules. The most informative frescoes are detected by means of two criteria and stored. The stored frescoes are considered as a human-like descritption of the robot's route and could be used by the robot to retrieve its route to its starting point.

The Global Positioning System (GPS) is now attracting worldwide attention as a navigation sensor. One of the advantages of GPS is that people can find their position using a GPS receiver wherever they are except in obstructed environments. Existing GPS receivers do not work in an obstructed environment despite there being many potential applications. This paper shows the possibility of navigation using GPS technologies indoors in a blocked environment. The paper describes the pseudolite-based indoor navigation system that was developed and implemented at Seoul National University (SNU). The system, which uses GPS technologies, has some different characteristics to normal outdoor GPS systems. The differences cause some practical problems, such as near/far, time-tag, multipath and unknown transmission position problems, which need to be solved to implement the indoor navigation system. The paper introduces various methods for solving such problems. The paper then shows the experimental results and the system accuracy. The satisfactory experimental results show that the RMS static error is 1 mm(1σ) horizontally and 2 mm(1σ) vertically and the RMS dynamic error is 5·6 mm(1σ) horizontally and 15 mm(1σ) vertically. The paper also shows the result of a field application test in the ocean engineering basin of the Korea Research Institute of Ships & Ocean Engineering (KRISO).