Liang Zize

Regression convolutional network for vanishing point detection

This paper presents a detection method for estimation of vanishing point position with designed regression convolutional neural network. Due to the deep structures of convolutional networks, global high-level features are extracted from the whole image, which helps to locate the vanishing point. In this paper, we provide a new structure of regression neural network based on AlexNet. The structure consists of five convolutional layers, four fully connected layers, Tanh activation function and regression loss function. We feed the neural net with a small number of training dataset and the result proves that this method is adaptable. Compare to classical method, deep learning is more effective on blurred pictures and complex circumstances.

An extrinsic calibration method for binocular cameras and swing 2D laser rangefinder

In this paper a novel method is proposed to calibrate the extrinsic parameters of the swing 2D laser rangefinder (LRF) and binocular cameras system. Target object that need special processing is abandoned. Extrinsic calibration is completed with one observation of three planes that posted with checkboard-like posters. The procedure is simplified. Superior to the monocular vision, stereo vision can obtain more information of scene. 3D virtual points and lines are deduced by analyzing the intersected 3D planes and they are used as inputs to an optimization problem which estimating the relative pose between camera and LRF. Moreover, we build geometric model of swing LRF and estimate the two important parameters with Levenberg-Marquardt (LM) individually. Thus, the distortion of point cloud that caused by mechanical structure is corrected as pretreatment of relative pose calculation. Experimental results demonstrate the validity of the proposed calibration and correction methods.

Servo system to control Arc length for electro-gas welding equipment

Automatic welding equipment with arc length control for electrolytic aluminum plant is designed. The servo system is composed of AC servo motor, ball screw, electro-gas (EGW) power, base material to be welded and servo controller. The welding power is installed of numerical communication port of welding current, providing information of velocity error between welding pool and gun. MIMO model of welding process is simplified into a linear error amplified. Dynamics of AC servo motor and ball screw with external load are analyzed. And a list of parameters is provided for further controlled design. The system is essentially a multi-loop control system. The internal loop of PID controller is tuned first. And then the whole system with PD controlled is tuned, so that the overshoot (σ%) and stabilization time (ts) is comfortable. At last, with the help of Simulink, arc control performance test is conducted. It proves that the system can keep arc length almost constant. This method is practical and easy to be realized for industrial application.

Random based narrow space path planning for arm manipulation with the fixed orientation constraints

Arm manipulation is a skill for robot working in cluttered environment. In this work, we focus on arm manipulation in narrow space with the end-effector to be a certain orientation. In this paper, we present a random, smooth path planning algorithm to avoid the obstacles considering every link of the arm, moving the arm from the starting point to the destination. Our planning algorithm can make the path similar in different planning process because the algorithm is end-point oriented. The simulation result validates the effectiveness of the algorithm.