Dwi Pebrianti

Real-time rotation invariant hand tracking using 3D data

Hand tracking is a common task in a gesture recognition system. Many techniques have been introduced to make successful hand tracking. In hand tracking system, most of previous works tracked the hand position using attached marker on hands. Several researchers have used a color image for skin color detection. However, using marker based need to attach marker on hands or wear gloves to make hand can be detected. When using color information, there is a need to extract many different skin colors. Furthermore, the lighting and background on the situation also need to be concerned to avoid a cluttered background that can affect the detection and tracking. This paper presents the real-time hand tracking using three dimensional (3D) data. This 3D data is coming from the Kinect sensor, which is working in real-time. 3D data from Kinect sensor is depth image data which can be used to detect and track the motion of the hand. This paper proposes hand tracking method using a hand tracker algorithm released by NiTE, hands segmentation method, hand contour detection and center of palm detection. The hands segmentation method consists of the ROI of the hands area and background subtraction. The propose hand tracking algorithm is rotation invariant, since it can detect and track various rotations of hand. Additionally, it also can remove unwanted object (noise) that also moving parallelly with the hands position.

Study on 3D scene reconstruction in robot navigation using stereo vision

In this paper, a 3D scene reconstruction by using stereo vision is presented. Stereo camera parameters from the camera calibration process and disparity map are two important parameters to obtain an accurate result for the 3D reconstruction. 3D reconstruction process generates the coordinates of world points (point cloud). From the information of the point cloud, the interested object is segmented out. Additionally, the noises left in the image is eliminated. The experimental result obtained shows that only the background and the interested object from the overall scene appeared in the processed image. The center of gravity is also determined as the reference value for the robot navigation. An estimation error model is also introduced in this study to increase the accuracy of the distance measurement by using the developed stereo vision system from the mobile robot. This experiment provides a foundation for navigation and object tracking for a mobile robot.

Performance of various speckle reduction filters on Synthetic Aperture Radar image

Synthetic Aperture Radar (SAR) image with its advantages, becoming popular than the optical image. However, the speckle in causes difficulties in the interpretation and analysis during image processing. Thus, before the SAR images are used, speckle noise reduction is necessary. The ideal speckle filter has the main goal of reducing speckle noise without losing the information, content, and preserve the edges and features. Various noise filters have been designed for different purposes and different capacities. In this study, we discuss four filters, namely Lee, Frost, Median and Mean filter. We are analyzing quality parameter and comparing statistic performance of Lee, Frost, Mean and Median filters for SAR sample data. The results show MSE, PSNR, SNR, and AD value that generate by Frost filter performs better than the other filter. And from visual interpretation of the de-speckle image that filtered with Frost filter, show sharpen edge and preserved texture.

Fault detection and identification in Quadrotor system (Quadrotor robot)

Fault Detection and Identification (FDI) monitor, identify, and pinpoint the type and location of system fault in a complex multiple input multiple output (MIMO) non-linear system. A Quadrotor robot is used to represent a complex system in this study. The aim of the research is to construct and design a Fault Detection and Isolation algorithm. This dynamic model is based on the first principles of the Quadrotor: Propeller model and its force as well as moments generation. The Quadrotor controller is designed such that it can be controlled using both the attitude control (inner loop) and position control (outer loop). PD controller used the Phi, Theta, Psi, x, y and z as a reference to adjust the attitude and position of the Quadrotor. The proposed method for the fault identification is a hybrid technique which combined both the Kalman filter and Artificial Neural Network (ANN). Kalman filter recognized data from the system sensors and can indicate the fault of the system in the sensor reading. Error prediction is based on the fault magnitude and the time occurrence of fault. The information will then be fed to Artificial Neural Network (ANN), which consist of a bank of parameter estimation that generates the failure state. This Artificial Neural Network (ANN) is an algorithm that is used to determine the type of fault and the severity level as well as isolate the fault from the system. The ANN is designed based on the back-propagation technique so that it can be trained to generate output based on the data. Based on the result comparison of the residual signal before filter and after filter, the algorithm of FDI is able to identify parts of the system that experience failure and the fault can be solved immediately allowing the Quadrotor to be back to its normal operation. It is also capable to acknowledge the user on the parts of the system which experienced failure and can provide user with the best instructions or solutions for the situation. It is also capable to cater a safe landing.

Comparison of fuzzy filters on Synthetic Aperture Radar image

The Synthetic Aperture Radar (SAR) image with its advantages is becoming more popular than the optical image in earth observation in using the remote-sensing techniques. However, the speckle noise that occurs in the SAR image causes difficulties in image interpretation. Thus, speckle noise reduction needs prepossessing procedure prior to the use of the SAR images. This study is done by proposed fuzzy filters that utilize SAR data. From the comparison, the combination of Frost-Triangular Moving Average (TMAV) has the best performance in the ability to reduce speckle noise than other filters. This filter improved the Frost filter performance for speckle noise reduction parameters measurement, shows that 13.41% for Equivalent Number of Looks (ENL) and 6.07% for Speckle Index (SI). While Frost-Asymmetric Triangular Moving Average (ATMAV) has a relatively good performance for preserved texture. This filter improved the texture parameters such as Standard Deviation improved 4.33% and improved Variance for 8.46%. However, for the Mean parameters, Frost-Triangular Median Center (TMED) combination has the best performance compared to other filters, which improved the mean value for 7.10%. The comparative study it has been verified that the fuzzy approach has the robustness in the reduction of speckle noise and preserving the texture when applied in SAR image.

Computer vision-based hand deviation exercise for rehabilitation

Computerized monitoring of the home based rehabilitation exercise has many benefits and it has attracted considerable interest among the computer vision community. Nowadays, many rehabilitation systems are proposed, most of the targeted disability is for stroke patient. Some of patient or user just wants to take certain part for rehabilitation. Therefore, this paper is focusing on hand rehabilitation system. The importance of the rehabilitation system is to implement the specific exercise for the specific requirements of the patients that needs rehabilitation therapy. This paper presents the specific hand rehabilitation system using computer vision method. The specific hand rehabilitation implemented in this system is a hand deviation exercise. This exercise is benefited to improve the mobility of the hand and reduce the pain. The hand tracking and finger detection method are used in this hand rehabilitation system. The result of the exercise can be used as a training data for the analysis of the injured hand recovery and healing process.

Finger application using K-Curvature method and Kinect sensor in real-time

Gesture is one of the important aspects of human interaction and also in the context of human computer interaction. Gesture recognition is the mathematical interpretation of a human motion by a computing device. It is often used hand gestures for input commands in personal computers. By recognizing the hand gesture as input, it allows the user to access the computer interactively and makes interaction more natural. This paper presents a finger detection application by using Kinect. Kinect is a depth sensor that is an effective device to capture the gesture in real-time. To detect and recognize the fingertips, it needs to extract the detail of the captured hand image using image processing methods. In this paper, the proposed method is to detect and recognize the fingertips by using the K-Curvature algorithm. Finally, the finger counting application is applied and the proposed method is discussed at the end of this paper. The results obtained from the experiment show that the acceptable average accuracy for the fingertips detection is 73.7% and the average processing time is 15.73 ms. By considering this result, the application of the proposed method can be extended to the hand rehabilitation system.

Motion Tracker Based Wheeled Mobile Robot System Identification and Controller Design

This project deals with the mathematical modelling and controller design for autonomous Wheeled Mobile Robot (WMR) by using motion tracking system. The mobile robot vehicle has two driving wheels and the angular speed of the two wheels is the controlled variable. Three reflected markers are attached on a robot to form a 3D rigid body. Motion tracker will track the 3D rigid body in terms of x, y position and orientation θ. Mathematical modelling which is a set of Multi Input Single Output model is done by using System Identification Toolbox in Matlab. Three different controller namely Proportional (P), Proportional Differential (PD) and Proportional Integral Differential (PID) controller are designed for this WMR. The mathematical model obtained from the System Identification has about 95% accuracy. In controller performance, the result shows that P, PD and PID controller have no overshoot for the forward movement. However, the percent overshoot of P, PD and PID controller when the robot is turning on side direction are around 51%, 63% and 48%, respectively. Additionally, the steady state error for all controllers is 0%.

Waypoint Navigation of Quad-rotor MAV Using Fuzzy-PID Control

Quad-rotor Micro Aerial Vehicle (MAV) is a multi-rotor MAV with four propellers which propel the MAV up to the air and move around. It has high maneuverability to move around, such as roll, pitch and yaw movements. However, line of sight and radio control effective range are the major limitation for the MAVs which significantly shorten the travel distance. Therefore, we proposed a waypoint navigation quad-rotor MAV based on Fuzzy-PID controller in this paper. User can set mission with multiple waypoint and the Fuzzy-PID controller will control MAV autonomously moving along the waypoint to the desired position without remotely controlled by radio control and guidance of pilot. The results show Fuzzy-PID controller is capable to control MAV to move to the desired position with high accuracy. As the conclusion, Fuzzy-PID controller is successfully designed for waypoint navigation in quad-rotor MAV. The result shows that the overshoot percentage (%OS) of the designed Fuzzy-PID controller for x position is 2.17% while y position is 0.93%. Additionally, the steady-state error for x position and y position are 0.54% and 0.56% respectively. Therefore, the performance of Fuzzy-PID controller is better than PID controller.

SKF-Based Image Template Matching for Distance Measurement by Using Stereo Vision

In this paper, a novel image template matching approach to tackle distance measurement problem has been proposed. There are many conventional algorithms to increase the accuracy of distance measurement as reported in the literature such as Semi-global algorithm to produce the disparity map. Meanwhile, in this paper, the reverse engineering technique had been implemented to get the correct depth value by applying the image template matching method as reference for the distance measurement. The traditional algorithm to solve image matching problem take a lot of memory and computational time. Therefore, image matching problem can be considered to optimization problem and can be solved precisely. The search of the image template has been performed exhaustively by using Simulated Kalman Filter (SKF) algorithm. The experiment is conducted with a set of images taken by using stereo vision system. Experimental results show the accuracy of the distance measurement by using stereo camera, after applying (1) the estimate error model, (2) SKF and (3) PSO algorithm are 89.95%, 96.09%, 95.29% and 58.51% respectively. The limitation of estimate error model that it can only be applied into the same setup of the experiment, environment, parameters of the camera and acquired images. Instead, the proposed algorithm which is SKF can be applied to original image and image under the vision problems like illumination and partially occluded. The SKF algorithm shows more robust, more efficient and more accurate to solve the distance measurement problem.