Rui Zhai

Bio: Rui Zhai is an academic researcher from Chinese Academy of Sciences. The author has contributed to research in topics: Software rendering & Real-time rendering. The author has an hindex of 4, co-authored 6 publications receiving 41 citations.

A Method of Vision Aided GNSS Positioning Using Semantic Information in Complex Urban Environment

Abstract: High-precision localization through multi-sensor fusion has become a popular research direction in unmanned driving. However, most previous studies have performed optimally only in open-sky conditions; therefore, high-precision localization in complex urban environments required an urgent solution. The complex urban environments employed in this study include dynamic environments, which result in limited visual localization performance, and highly occluded environments, which yield limited global navigation satellite system (GNSS) performance. In order to provide high-precision localization in these environments, we propose a vision-aided GNSS positioning method using semantic information by integrating stereo cameras and GNSS into a loosely coupled navigation system. To suppress the effect of dynamic objects on visual positioning accuracy, we propose a dynamic-simultaneous localization and mapping (Dynamic-SLAM) algorithm to extract semantic information from images using a deep learning framework. For the GPS-challenged environment, we propose a semantic-based dynamic adaptive Kalman filtering fusion (S-AKF) algorithm to develop vision aided GNSS and achieve stable and high-precision positioning. Experiments were carried out in GNSS-challenged environments using the open-source KITTI dataset to evaluate the performance of the proposed algorithm. The results indicate that the dynamic-SLAM algorithm improved the performance of the visual localization algorithm and effectively suppressed the error spread of the visual localization algorithm. Additionally, after vision was integrated, the loosely-coupled navigation system achieved continuous high-accuracy positioning in GNSS-challenged environments.

Challenge of multi-camera tracking

Abstract: Multi-camera tracking is quite different from single camera tracking in mathematical principles and application scenarios, and it faces new technology and system architecture challenges. The existing theories and algorithms used in object matching, cameras calibration and topology estimation, and information fusion have been reviewed and show that the integrated application of multi techniques and multi theories is the key to solve the technology challenges. The distributed architectures of multi-camera tracking system based on camera processor and based on object agent have been compared and show that improving the computation ability of cameras and reducing the functions of control center is the key to solve the architecture challenges.

Large scale terrain visualization using the restricted quadtree triangulation

Abstract: Real-time rendering of triangulated surfaces has attracted growing interest in the last few years. However, interactive visualization of very large scale grid digital elevation models is still difficult. The graphics load must be controlled by adaptive surface triangulation and by taking advantage of different levels of detail. Furthermore, management of the visible scene requires efficient access to the terrain database. We describe an all-in-one visualization system which integrates adaptive triangulation, dynamic scene management and spatial data handling. The triangulation model is based on the restricted quadtree triangulation. Furthermore, we present new algorithms of restricted quadtree triangulation. These include among others exact error approximation, progressive meshing, performance enhancements and spatial access.

Stability Analysis of the Modified Levenberg–Marquardt Algorithm for the Artificial Neural Network Training

Abstract: The Levenberg–Marquardt and Newton are two algorithms that use the Hessian for the artificial neural network learning. In this article, we propose a modified Levenberg–Marquardt algorithm for the artificial neural network learning containing the training and testing stages. The modified Levenberg–Marquardt algorithm is based on the Levenberg–Marquardt and Newton algorithms but with the following two differences to assure the error stability and weights boundedness: 1) there is a singularity point in the learning rates of the Levenberg–Marquardt and Newton algorithms, while there is not a singularity point in the learning rate of the modified Levenberg–Marquardt algorithm and 2) the Levenberg–Marquardt and Newton algorithms have three different learning rates, while the modified Levenberg–Marquardt algorithm only has one learning rate. The error stability and weights boundedness of the modified Levenberg–Marquardt algorithm are assured based on the Lyapunov technique. We compare the artificial neural network learning with the modified Levenberg–Marquardt, Levenberg–Marquardt, Newton, and stable gradient algorithms for the learning of the electric and brain signals data set.

A comparison between the application of empirical and ANN methods for estimation of daily global solar radiation in Iran

Abstract: The present study generally aims to provide a comparison between the performance and suitability of different types of models for estimation of daily global solar radiation in Iran, based on duration of sunshine hours and diurnal air temperature. These models consist of empirical, ordinary ANN, and ANN models coupled with genetic algorithm (so called coupled ANN models). The models’ performance was evaluated and compared based on the error statistics root mean squared error (RMSE), mean bias error (MBE), and coefficient of determination (R2). The empirical models (median of R2, MBE, RMSE for AP 0.93, 37.0, and 179.3 J/cm2/day) could generally perform much better than the ordinary ANN models (median of R2, MBE, RMSE for MLP(n) 0.90, 55.7, and 243.5 J/cm2/day). The performance of the ordinary ANN models was improved considerably after being coupled by genetic algorithm (median of R2, MBE, RMSE for MLP-GA(n) 0.92, 38.4, and 185.5 J/cm2/day), making them the most accurate models at most of the stations studied. However, the difference between the overall performances of these coupled ANN models and empirical ones was slight. Lastly, despite the coupled ANN models had relatively better accuracy compared to the empirical ones, when taking different metrics such as the required processing time, skill, and equipment into account, the empirical models appear to be the most suitable models for estimation of daily global solar radiation in Iran.

Pedestrian Models for Autonomous Driving Part I: Low-Level Models, From Sensing to Tracking

Abstract: Autonomous vehicles (AVs) must share space with pedestrians, both in carriageway cases such as cars at pedestrian crossings and off-carriageway cases such as delivery vehicles navigating through crowds on pedestrianized high-streets. Unlike static obstacles, pedestrians are active agents with complex, interactive motions. Planning AV actions in the presence of pedestrians thus requires modelling of their probable future behavior as well as detecting and tracking them. This narrative review article is Part I of a pair, together surveying the current technology stack involved in this process, organising recent research into a hierarchical taxonomy ranging from low-level image detection to high-level psychology models, from the perspective of an AV designer. This self-contained Part I covers the lower levels of this stack, from sensing, through detection and recognition, up to tracking of pedestrians. Technologies at these levels are found to be mature and available as foundations for use in high-level systems, such as behavior modelling, prediction and interaction control.

Local Levenberg-Marquardt Algorithm for Learning Feedforwad Neural Networks

Abstract: Abstract This paper presents a local modification of the Levenberg-Marquardt algorithm (LM). First, the mathematical basics of the classic LM method are shown. The classic LM algorithm is very efficient for learning small neural networks. For bigger neural networks, whose computational complexity grows significantly, it makes this method practically inefficient. In order to overcome this limitation, local modification of the LM is introduced in this paper. The main goal of this paper is to develop a more complexity efficient modification of the LM method by using a local computation. The introduced modification has been tested on the following benchmarks: the function approximation and classification problems. The obtained results have been compared to the classic LM method performance. The paper shows that the local modification of the LM method significantly improves the algorithm’s performance for bigger networks. Several possible proposals for future works are suggested.