Zhiyu Xiang

Polarization-Based Water Hazards Detection for Autonomous Off-road Navigation

A polarization-based method for water hazards detection is presented. The concept of polarization is introduced to computer vision to detect water hazards for autonomous off-road navigation. This method is based on the physical principle that the light reflected from water surface is partial linearly polarized and the polarization phases of them are more similar than those from the scenes around. Water hazards can be detected by comparison of polarization degree and similarity of the polarization phases. Experiments show that the method has good performance in water detection in complex natural backgrounds, especially when there is vegetation reflected in the water. This method has significant complementary advantages with respect to existing techniques, is computationally efficient, and can be easily implemented with existing imaging technology.

Multi-Feature Fusion Based Outdoor Water Hazards Detection

Water hazards such as ponds usually threaten the unmanned ground vehicle autonomous off-road navigation. The water hazards detection is a great challenge when using the machine vision. The method of multi-feature fusion is applied here for this kind of detection. By using the color cameras, we extract the features of brightness and texture from the non-reflection regions. Further more, a new stereo vision based method is also proposed to get the height and the distance information of the reflection regions, and then the corresponding water regions can be acquired. Finally all the features are fused together and the accurate water regions from the images can be detected. The experimental results demonstrate the efficiency of this method.

Design and Calibration of a Fast 3D Scanning LADAR

A fast 3D scanning laser radar is designed and calibrated. A set of mechanical equipment driven by a step motor is designed to rotate the scanning plane of a 2D laser range finder to realize 3D scanning. A micro-controller system is embedded for the scanning control and synchronization with the host computer. Measurement data are acquired by a self-developed high-speed data transmission board. Besides the inherent measurement error of the 2D laser range finder, the error of the resulting 3D LADAR system is mainly caused by two reasons: 1) the unknown offset between the axis of the pitching motion and the actual origin of the 2D laser range finder; 2) the initial attitude error. Based on the system measurement model, a least sum of squared error method is adopted to calibrate the system. The experimental results show that the resulting system has high scanning speed, high data accuracy and therefore is a cost effective means to acquire high quality 3D data

An adaptive fast search algorithm for block motion estimation in H.264

Motion estimation is an important issue in H.264 video coding systems because it occupies a large amount of encoding time. In this paper, a novel search algorithm which utilizes an adaptive hexagon and small diamond search (AHSDS) is proposed to enhance search speed. The search pattern is chosen according to the motion strength of the current block. When the block is in active motion, the hexagon search provides an efficient search means; when the block is inactive, the small diamond search is adopted. Simulation results showed that our approach can speed up the search process with little effect on distortion performance compared with other adaptive approaches.

Stereo vision based SLAM in outdoor environments

Simultaneous localization and mapping is a well studied problem as it is considered by many to be an essential capability for autonomous robots. In this paper, we present an algorithm using a stereo camera based on Rao-Blackwelltion particle filter, It can realize three dimensional stereo vision SLAM for mobile robot in unknown outdoor environments. Firstly, we determine the initial motion estimation between two adjacent frames through the multiple view geometry utilizing the matched SIFT point pairs. Then, the 3D positions of landmarks are constructed directly through triangulation methodology. With accurate data association, the cameras state and the landmark positions are updated recursively. Finally, an efficient particle resamping algorithm MPR (the modified particle resampling algorithm) is proposed to deal with the degeneracy of particles.

Design of an enhanced visual odometry by building and matching compressive panoramic landmarks online

Efficient and precise localization is a prerequisite for the intelligent navigation of mobile robots. Traditional visual localization systems, such as visual odometry (VO) and simultaneous localization and mapping (SLAM), suffer from two shortcomings: a drift problem caused by accumulated localization error, and erroneous motion estimation due to illumination variation and moving objects. In this paper, we propose an enhanced VO by introducing a panoramic camera into the traditional stereo-only VO system. Benefiting from the 360° field of view, the panoramic camera is responsible for three tasks: (1) detecting road junctions and building a landmark library online; (2) correcting the robot’s position when the landmarks are revisited with any orientation; (3) working as a panoramic compass when the stereo VO cannot provide reliable positioning results. To use the large-sized panoramic images efficiently, the concept of compressed sensing is introduced into the solution and an adaptive compressive feature is presented. Combined with our previous two-stage local binocular bundle adjustment (TLBBA) stereo VO, the new system can obtain reliable positioning results in quasi-real time. Experimental results of challenging long-range tests show that our enhanced VO is much more accurate and robust than the traditional VO, thanks to the compressive panoramic landmarks built online.

Polarization-based water hazards detection for autonomous off-road navigation

A polarization-based method for water hazards detection is presented. The concept of polarization is introduced to computer vision to detect water hazards for autonomous off-road navigation. This method is based on the physical principle that the light reflected from water surface is partial linearly polarized and the polarization phases of them are more similar than those from the scenes around. Water hazards can be detected by comparison of polarization degree and similarity of the polarization phases. Experiments show that the method has good performance in water detection in complex natural backgrounds, especially when there is vegetation reflected in the water. This method has significant complementary advantages with respect to existing techniques, is computationally efficient, and can be easily implemented with existing imaging technology.

Environmental perception: an application of multi-sensor data fusion to autonomous off-road navigation

Environmental perception is one of the most difficult problems for off-road autonomous vehicles. Due to the variety and complexity of off-road environments, the information from any single sensor is not enough for safe and efficient vehicle navigation. Employing more sensors can greatly improve the vehicles perceptive capability. This paper describes a multi-sensor data fusion system for off-road autonomous vehicles. The system acquires data from one camera, four laser range finders, one radar, and several ultrasonic sensors. A hierarchical structure is used to organize the sensors from feature level to high fusion level. Dempster-Shafer evidence theory is adopted to decide the classification of each grid in the fusion map. A weighted evidence combination rule is proposed and implemented to improve the decision results under high conflicting circumstance. The experimental results showed the validity of our method.

Object tracking with 3D LIDAR via multi-task sparse learning

Moving object tracking is a fundamental task for autonomous vehicles operating in urban areas. In this paper, a novel sparse learning based object tracking algorithm utilizing 3D LIDAR data is proposed. The 3D point clouds acquired from HDL-64E 3D LIDAR are first resampled on a virtual image plane, where the hypothesis of the targets is generated under the particle filtering framework. Four complementary features, i.e., normal orientation, depth, LBP and HOG, are extracted on each particle to describe the appearance of the candidates. Then a multi-task multi-cue sparse learning algorithm is employed to select the best candidate and realize the tracking of the object. To improve the robustness of the algorithm, the sparse learning framework is further enhanced by a specifically designed background filtering and occlusion detection mechanism. The experiments carried out on KITTI benchmark show promising object tracking performance, especially when handling complex tracking situations such as occlusion and posture change.

An Efficient Monte Carlo Method for Mobile Robot Localization

Traditional Kalman filter or extended Kalman filter has been used broadly for mobile robot localization. However in some circumstances, its prior Gaussian hypothesis becomes unacceptable and limits the localization precision. For this reason, Monte Carlo method is used for the robot localization. In this paper, the algorithm implementation is advanced after introducing the basic theory of Monte Carlo method. According to the characteristics of robot movement, a new resample method is presented based on adapting the sample size and their particles space distribution. Finally experiments confirm the advantage of boosting accuracy and convergence speed about this idea