My Ha Le

3D motion estimation based on pitch and azimuth from respective camera and laser rangefinder sensing

This paper proposes a new method to estimate the 3D motion of a vehicle based on car-like structured motion model using an omnidirectional camera and a laser rangefinder. In recent years, motion estimation using vision sensor has improved by assuming planar motion in most conventional research to reduce requirement parameters and computational cost. However, for real applications in environment of outdoor terrain, the motion does not satisfy this condition. In contrast, our proposed method uses one corresponding image point and motion orientation to estimate the vehicle motion in 3D. In order to reduce requirement parameters for speedup computational systems, the vehicle moves under car-like structured motion model assumption. The system consists of a camera and a laser rangefinder mounted on the vehicle. The laser rangefinder is used to estimate motion orientation and absolute translation of the vehicle. An omnidirectional image-based one-point correspondence is used for combining with motion orientation and absolute translation to estimate rotation components of yaw, pitch angles and three translation components of Tx, Ty, and Tz. Real experiments in sloping terrain demonstrate the accuracy of vehicle localization estimation using the proposed method. The error at the end of travel position of our method, one-point RANSAC are 1.1%, 5.1%, respectively.

Vehicle localization using omnidirectional camera with GPS supporting in wide urban area

This paper proposes a method for long-range vehicle localization using fusion of omnidirectional camera and Global Positioning System (GPS) in wide urban environments. The main contributions are twofold: first, the positions estimated by visual sensor overcome the motion blur effects. The motion constrains of successive frames are obtained accurately under various scene structures and conditions. Second, the cumulative errors of visual odometry system are solved completely based on the fusion of local (visual odometry) and global positioning system. The visual odometry can yield the correct local position at short distance of movements but it will accumulate errors overtime, on the contrary, the GPS can yields the correct global positions but the local positions may be drifted. Moreover, the signals received from satellites are affected by multi-path and forward diffraction then the position errors increase when vehicles move in dense building regions or jump/miss in tunnels. To utilize the advantages of two sensors, the position information should be evaluated before fusion by Extended Kalman Filter (EKF) framework. This multiple sensor system can also compensate each other in the case of losing one of two. The simulation results demonstrate the accuracy of vehicle positions in long-range movements.

Robust human detection using multiple scale of cell based histogram of oriented gradients and adaboost learning

Human detection is an important task in many applications such as intelligent transport systems, surveillance systems, automatic human assistance systems, image retrieval, and so on. This paper proposes a multiple scale of cell based Histogram of Oriented Gradients (HOG) features description for human detection system. Using these proposed feature descriptors, a robust system is developed according to decision tree structure of boosting algorithm. In this system, the integral image based method is utilized to compute feature descriptors rapidly, and then cascade classifiers are taken into account to reduce computational cost. The experiments were performed on INRIAs database and our own database, which includes samples in several different sizes. The experiment results showed that our proposed method produce high performance with lower false positive and higher recall rate than the standard HOG features description. This method is also efficient with different resolution and gesture poses under a variety of backgrounds, lighting, as well as individual human in crowds, and partial occlusions.

3D scene reconstruction enhancement method based on automatic context analysis and convex optimization

Abstract This paper proposes a method for increasing accuracy of the scene model generation. Reconstruction of 3D scene model is strongly affected by moving objects both artificial and natural. In the proposed method context analysis is applied as a pre-filtering operation used to detect and remove objects which could negatively affect reconstruction process. Additionally, robust global rotation constraints are computed based on correspondence of image pairs. These constraints are fed to the model generation procedure. Finally, in contrast with using only canonical bundle adjustment which yields unstable structure in critical configuration and local minima, the proposed method utilized known-rotation frame work to compute the initial guess for bundle adjustment process which overcomes the drawback above. Moreover, the patch-based multi-view stereopsis is applied to upgrade the reconstructed structure. The simulation results demonstrate the tidy of structures reconstructed by this method from scene images in outdoor environment.

Motion Estimation Based on Two Corresponding Points and Angular Deviation Optimization

Recently, there have been several studies on vision-based motion estimation under a supposition that planar motion follows a nonholonomic constraint. This allows reducing computational time. However, the vehicle motion in an outdoor environment does not accept this assumption. This paper presents a method for estimating the vision-based 3-D motion of a vehicle with several parts as follows. First, the Ackermann steering model is applied to reduce constraint parameters of the 3-D motion. In difference to the previous contribution, the proposed approach requires only two corresponding points of consecutive images to estimate the vehicle motion. Second, motion parameters are extracted based on a closed-form solution on geometric constraints. Third, the estimation approach applies the bundle adjustment-based quasiconvex optimization. This task aims to take into account advantage of omnidirectional vision-based features for reducing errors. The omnidirectional vision supports for landmarks tracking in long travel and large rotation, which is appropriate for a bundle adjustment technique. Evaluated results show that the proposed method is applicable in the practical condition of outdoor environments.

Automatic vehicle identification by plate recognition for intelligent transportation system applications

Automatic vehicle identification is a very crucial and inevitable task in intelligent traffic systems. In this paper, initially, a Hue-Saturation-Intensity (HSI) color model is adopted to select automatically statistical threshold value for detecting candidate regions. The proposed method focuses are on the implementation of a method to detect candidate regions when vehicle bodies and license plate (LP) have similar color based on characteristics of color. Tilt correction in horizontal direction by the least square fitting with perpendicular offsets (LSFPO) is proposed and implemented for estimating rotation angle of the LP region. Then the whole image is rotated for tilt correction in horizontal direction by this angle. Tilt correction in vertical direction by reorientation of the titled LP candidate through inverse affine transformation is proposed and implemented for removing shear from the LP candidates. Finally, statistical based template matching technique is used for recognition of Korean plate characters. Various LP images are used with a variety of conditions to test the proposed method and results are presented to prove its effectiveness.

Building face reconstruction from sparse view of monocular camera

This paper proposes a method for building detection and 3D reconstruction of building face from sparse view of monocular camera. According to this method, building faces are detected by using color, straight line, edge and vanishing point. In the next step, building faces from multi view are extracted. Point clouds of building face are obtained from triangulation step. The building faces are reconstructed by plane fitting afterward. The simulation results will demonstrate the effectiveness of this method.

Building detection and 3D reconstruction from two-view of monocular camera

This paper proposes a method for building detection and 3D reconstruction from two-view by using monocular system. According to this method, building faces are detected by using color, straight line, edge and vanishing point. In the next step, invariant features are extracted and matching to find fundamental matrix. Three-dimension reconstruction of building is implemented based on camera matrixes which are computed from fundamental matrix and camera calibration parameters (essential matrix). The true dimension of building will be obtained if assume the baseline of monocular system is known. The simulation results will demonstrate the effectiveness of this method.

Boosting Discriminative Models for Activity Detection Using Local Feature Descriptors

This paper presents a method for daily living activity prediction based on boosting discriminative models. The system consists of several steps. First, local feature descriptors are extracted from multiple scales of the sequent images. In this experiment, the basic feature descriptors based on HOG, HOF, MBH are considered to process. Second, local features based BoW descriptors are studied to construct feature vectors, which are then fed to classification machine. The BoW feature extraction is a pre-processing step, which is utilized to avoid strong correlation data, and to distinguish feature properties for uniform data for classification machine. Third, a discriminative model is constructed using the BoW features, which is based on the individual local descriptor. Sequentially, final decision of action classes is done by the classifier using boosting discriminative models. Different to previous contributions, the sequent-overlap frames are considered to convolute and infer action classes instead of an individual set of frames is used for prediction. An advantage of boosting is that it supports to construct a strong classifier based on a set of weak classifiers associated with appropriate weights to obtain results in high performance. The method is successfully tested on some standard databases.

Enhancing 3D Scene Models Based on Automatic Context Analysis and Optimization Algorithm

This paper proposes a method for enhancing accuracy of scene model. The main contributions are threefold: first, the contex of the scene images are analyzed. Some objects which may have negative effect should be removed. For instance, the sky often appears as backgroud and moving objects appear in most of scene images. They are also one of reasons that causes the outliers. Second, the global rotations of images are computed based on correspondence between pair-wise images. These constraints are fed to the point clouds generation procedure. Third, in contrast with using only canonical bundle adjustment which yields unstable structure in small baseline geometry and local minima, the proposed method utilized known-rotation framework to compute the initial guess for bundle adjustment process. The patch-based multi-view stereopsis is applied to upgrade the reconstructed structure. The simulation results demonstrate the accuracy of structures by this method from scene images in outdoor environment.