Shih-Tseng Lee

Human face detection with neural networks and the DIRECT algorithm

Based on Rowley’s approach, this article proposes a new architecture that uses a specific optimization technique, the DIRECT (DIviding RECTangle) algorithm, to improve the efficiency of face detection in images. The system consists of two main parts: a neural network-based face detection arbitrator, and a search strategy based on an integer-handling DIRECT algorithm. By the architecture, the number of arbitrations is dramatically reduced, and human faces, if they are present in an image, are not restricted to predetermined resolutions and aspect ratios. Experimental results show that the proposed architecture is efficient in terms of both speed and robustness.

3D registration of human face using evolutionary computation and Kriging interpolation

This paper proposes a fast and robust 3D human face geometric data registration strategy dedicated for image-guided medical applications. The registration scheme is composed of a coarse transformation stage and a fine-tuning stage. In the first stage, fuzzy c-mean is used to reduce the data amount of template 3D image, and evolutionary computation is implemented to find optimal initial pose for the Iterative Closest Point plus k-dimensional (KD) tree scheme. In the second stage, the huge reference image data are replaced by a Kriging model. The time-consuming search for corresponding points in evaluating the degree of misalignment is substituted by projecting the points in the template image onto the model. To illustrate the validity and applicability of the proposed approach, a problem composed of 174 635 points reference image and an 11 280 points template image is demonstrated. Computational results show that our approach accelerates the registration process from 1361.28 seconds to 432.85 seconds when compared with the conventional ICP plus K-D tree scheme, while the average misalignment reduces from 11.35 mm to 2.33 mm.

Construction of Fuzzy Map for Autonomous Mobile Robots Based on Fuzzy Confidence Model

This paper presents the use of fuzzy models to explicitly consider sensor uncertainty and finite resolution in solving the SLAM (simultaneous localization and mapping) problem for autonomous mobile robots. The approach establishes fuzzy confidence models in describing occupied obstacles and available space. The problem is transformed into an optimization task of minimizing the alignment error between newly scanned local fuzzy maps and selected parts of a developing global fuzzy map. In aligning local fuzzy maps into a global fuzzy map, we developed a prediction strategy to crop the most potential part from the sensed local fuzzy maps to be overlapped with the global fuzzy map. A mobile vehicle equipped with a laser range finder, the Hokuyo URG-04LX, is used to demonstrate the procedure of fuzzy map building. Experimental results show that the proposed architecture is effective in generating a comprehensive global fuzzy map, which is suitable for both human comprehension and path design during real-time navigation.

Efficient Stereo Matching with Decoupled Dissimilarity Measure Using Successive Weighted Summation

Developing matching algorithms from stereo image pairs to obtain correct disparity maps for 3D reconstruction has been the focus of intensive research. A constant computational complexity algorithm to calculate dissimilarity aggregation in assessing disparity based on separable successive weighted summation (SWS) among horizontal and vertical directions was proposed but still not satisfactory. This paper presents a novel method which enables decoupled dissimilarity measure in the aggregation, further improving the accuracy and robustness of stereo correspondence. The aggregated cost is also used to refine disparities based on a local curve-fitting procedure. According to our experimental results on Middlebury benchmark evaluation, the proposed approach has comparable performance when compared with the selected state-of-the-art algorithms and has the lowest mismatch rate. Besides, the refinement procedure is shown to be capable of preserving object boundaries and depth discontinuities while smoothing out disparity maps.

Uniformly spaced 3D modeling of human face from two images using parallel particle swarm optimization

This paper proposes a scheme for finding the correspondence between uniformly spaced locations on the images of human face captured from different viewpoints at the same instant. The correspondence is dedicated for 3D reconstruction to be used in the registration procedure for neurosurgery where the exposure to projectors must be seriously restricted. The approach utilizes structured light to enhance patterns on the images and is initialized with the scale-invariant feature transform (SIFT). Successive locations are found according to spatial order using a parallel version of the particle swarm optimization algorithm. Furthermore, false locations are singled out for correction by searching for outliers from fitted curves. Case studies show that the scheme is able to correctly generate 456 evenly spaced 3D coordinate points in 23 seconds from a single shot of projected human face using a PC with 2.66 GHz Intel Q9400 CPU and 4GB RAM.

Calibration of a dual-PTZ camera system for stereo vision

In this paper, we propose a calibration process for the intrinsic and extrinsic parameters of dual-PTZ camera systems. The calibration is based on a complete definition of six coordinate systems fixed at the image planes, and the pan and tilt rotation axes of the cameras. Misalignments between estimated and ideal coordinates of image corners are formed into cost values to be solved by the Nelder-Mead simplex optimization method. Experimental results show that the system is able to obtain 3D coordinates of objects with a consistent accuracy of 1 mm when the distance between the dual-PTZ camera set and the objects are from 0.9 to 1.1 meters.

Integration of a stereo vision system and a laser range finder for 3-D construction

In this article, we investigate the problem of integrating a binocular stereo vision system and a laser range finder to construct a 3-D map of the environment. The proposed scheme is realized by using the alignment parameters obtained in the 2-D map construction of the laser range finder for the 3-D data generated by the stereo vision system. The 2-D map alignment task is formulated as an optimization problem of minimizing the alignment errors between local maps and selected parts of the developing global map. The problem is then solved using the Simplex method. To increase the robustness of the searching process, multiple initial guesses are provided in the Simplex method. The performance of the proposed architecture is verified by experimental results from a mobile vehicle for obstacle avoidance.

Development of a visual compressive trackng system enhanced by adaptive boosting

This paper presents the development of a dynamic optical tracking system based on a PTZ camera platform to stably track any selected object and keep the targeted object image in the center of video images. The system is based on the Compressive Tracking algorithm which provides dimensionality reduction to achieve real-time tracking. To enhance reliability and avoid failure due to morphological variation, disappearing and blocking of the targeted objects on image frames, the developed tracking algorithm is enhanced with an Adaptive Boosting (Ada-Boost) evaluation strategy to improve the reliability of the weak Naive Bayes classifier. Besides, a heuristic threshold is applied to select and update object templates to avoid drifting of the templates. Furthermore, potential locations of target patches are searched evenly for based on the Halton sequence. Experiments on practical scenarios demonstrate the robustness and effectiveness of the proposed scheme.

Calibration of a dual-PTZ-camera system for stereo vision based on parallel particle swarm optimization method

This work investigates the calibration of a stereo vision system based on two PTZ (Pan-Tilt-Zoom) cameras. As the accuracy of the system depends not only on intrinsic parameters, but also on the geometric relationships between rotation axes of the cameras, the major concern is the development of an effective and systematic way to obtain these relationships. We derived a complete geometric model of the dual-PTZ-camera system and proposed a calibration procedure for the intrinsic and external parameters of the model. The calibration method is based on Zhang’s approach using an augmented checkerboard composed of eight small checkerboards, and is formulated as an optimization problem to be solved by an improved particle swarm optimization (PSO) method. Two Sony EVI-D70 PTZ cameras were used for the experiments. The root-mean-square errors (RMSE) of corner distances in the horizontal and vertical direction are 0.192 mm and 0.115 mm, respectively. The RMSE of overlapped points between the small checkerboards is 1.3958 mm.

Application of Real-time Single Camera SLAM Technology for Image- guided Targeting in Neurosurgery

In this paper, we propose an application of augmented reality technology for targeting tumors or anatomical structures inside the skull. The application is a combination of the technologies of MonoSLAM (Single Camera Simultaneous Localization and Mapping) and computer graphics. A stereo vision system is developed to construct geometric data of human face for registration with CT images. Reliability and accuracy of the application is enhanced by the use of fiduciary markers fixed to the skull. The MonoSLAM keeps track of the current location of the camera with respect to an augmented reality (AR) marker using the extended Kalman filter. The fiduciary markers provide reference when the AR marker is invisible to the camera. Relationship between the markers on the face and the augmented reality marker is obtained by a registration procedure by the stereo vision system and is updated on-line. A commercially available Android based tablet PC equipped with a 320×240 front-facing camera was used for implementation. The system is able to provide a live view of the patient overlaid by the solid models of tumors or anatomical structures, as well as the missing part of the tool inside the skull.