Article Preview
TopIntroduction
Construction cranes are one of the most important industrial vehicles for transporting goods. Some operators have to automatically move the cart slowly to the destination site, and the others carefully drive the crane while others watch the motion of the crane on the ground. There are two vital factors in the crane transporting system, jostling of the load and swinging of the cart (Hirata et al., 2007), which makes the overhead crane system difficult to be controlled. Tracking the load is the most important step in the entire process of working with cranes. The goal of this paper is to successfully keep track of the crane using cameras, insuring smooth and rapid transportation.
Other nonlinear control methods have been presented to solve managing problems in engineering. These control methods include the following: adaptive attitude control (Chen & Huang, 2009), sliding mode control (Lu et al., 2012), output feedback control (Wong et al., 2001), passivity-based control (Pisu & Serrani, 2007), fuzzy control (Zou et al., 2011). The AFSMC method (Chang & Lie, 2012) with adaptable sliding slopes using the encoders to observe the angle of the loads swing and the trolley’s position, combining the merits of SMC and model-free property of fuzzy logic control (FLC). Viet (2015) took it into consideration the spring and damper caused by payload and the cable. However, the vibration induced by trolley and other parts of crane could cause the displacement of the sensor and reduce data reliability. It is important and necessary for dynamic systems to apply visual tracking in many industrial fields. There are more and more researchers who have tried to apply image sensing to the angle at which construction crane’s sing they loads or have used tried to track an object using recent technology that observes visual cues, however it is hard to use it in real-time control of overhead crane (Xu et al., 2006; Celik & Kusetogulari, 2010; Liaw & Shirinzadeh, 2011; Kawai et al., 2009).Osumi et al. (2005) attempted to calculate 2D swing’s angle by two CCD cameras, however it is burdensome for the required calculations. Matsuo et al. (2004) put forward a PID+Q crane controller with visual feedback by building a video tracker. The crane controllers are put forward by Yoshida and Tsuzuki (2006) and Lee et al. (2014) using stereovision cameras as feedback sensors which were adopted on robots. Yang et al. (2017) proposed a novel correlation filter-based tracker with robust estimation of similarity transformation on the large displacements to tackle this challenging problem. However, in some complicated scenes, there is a fatal defect that may cause the object to be located inaccurately. Di et al. (2017) proposed a particle filter redetection-based tracking approach for accurate object localization. Nevertheless, the preceding methods were hard to use in complicated circumstances without devices capable of high rates of computations. Furthermore, these visual tracking methods require lots of sensing instruments.
A sparse measurement matrix is a method widely used when it comes to numerical issues. Cui et al. (2018) etc. employed the Sparse Measurement Matrix (DSMM) in convolutional network to reduce the sampling computational complexity and improve the CS reconstruction performance. Stefan et al. (2017) proposed the sparse measurement matrices in the paper dramatically simplify the circuit implementation and relax the signal swing requirement. Zhen et al. (2016) also presented to a novel compressive sensing method based on SVD sparse random measurement matrix in wireless sensor network enhance quality of reconstruction of compressive sensing signal. A sparse matrix is a type of special matrix in which non-zero elements account for a small percentage of all elements of the matrix. There are many elements equal to zero in a sparse Matrix, which is convenient for matrix calculation and preservation.
A simple but effective method to visually track objects is to capture the movement of the load of the overhead crane; this allows the controller to automatically operate in real time. The process used in this paper using a CCD video camera to observe the crane’s load. This camera also helps to make image processing time faster and more efficient during the entire process.
Firstly, a measurement matrix is used for extracting the details of the appearance model. Then, the load located in the foreground, along with the background are then turned into compressed samples which can be used when creating a sparse measurement matrix. Lastly, a Naïve Bayesian classifier is used to follow and predict the movement of crane’s and their loads which is ultimately uploaded on a compressed domain.