The Autopilot System For Vision-guided Agv

In last 30 years, Automatic Guided Vehicle (AGV) has developed into one of the largest research branches of manufacture and logistics system. And there is the trend of industrial development.It will turn to be an important and indispensable component of automation equipment in a modern enterprise. With the development of artificial intelligence, image processing technology and other fields, using computer vision to guide the AGV has become an important means of navigation, and also a research hotspot in recent years. The method gets the road image info through the camera and processes it to access to the road parameters, and then guide the AGV with it. Compared with the traditional guidance technology, vision-guided system is costing less, adaptable and flexible installation and so on. But existing vision-guided technology is not mature enough yet, mainly reflected in the stability and robustness of the deficiency.This paper designs a set of vision-guided vehicle AGV system solutions, which is optimized or innovated from multiple aspects of the existing methods. This vision guidance system designed to capture a single camera as input, combined with RFID, global map information and the local motor state, the output is accurate road parameters to go along with, as the final input of the control module.This study includes the following three aspects:1) Within the system, image processing module is the core module, while the highest consumption of calculating resources. In order to meet the real-time requirements, based on that the target arc has relatively large radius, a simplified Hough-line-basis Hough circle transform is presented. It reduces Hough circle transformation's parameter space to two-dimensional. In the condition that straight edge detection must be done, it will save a greatly extra computation time to calculate arc recognition. This method can meet the needs of real-time in this system.2) Since the image processing technology proposed in 1) has a single target, not include the calculation of complex features, there is not good anti-noise performance. To work with this problem, we propose to use some prior knowledge (the road width and curve radius, etc.) to enhance the accuracy of image processing. And through a global map and RFID signal response, the rough location of AGV can be achieved, and predict the geometry characteristic of targets, makes image recognition targeted to reduce false recognition rate and recognition rate of leakage.3) The error of the results is made of two major components, namely:a).the time gap between image acquisition and control instructions; b). Measurement error. In this paper, the prediction method is used, after analyze the physical process of moving AGV, the recorded history recognition parameters is used to calculate the least squares polynomial fitting, predict the approximation of the real parameters. Also this approximation statistical probability distribution-based particle filter is used to further reduce the error of parameter estimates. With the minimization of the error of input values to PID control module, the convergence speed of straight traveling control can be effectively shorten.The various components of the above algorithm operate as a whole, with mutual cooperation and the optimal use of their results. After verification of the real experiment scene, under the control of the visual guidance method proposed, AGV shows a stable and efficient implementation of the mandate of the prior arrangement, to project the initial request.