Implementation of Low Cost Vision Based Measurement System: Motion Analysis of Indoor Robot

Several measurement devices are available in the market to obtain the position, velocity, and trajectory of mobile robots. However in an indoor environment, it is expensive to build a reliable measurement system by assembling these devices. Some of the low cost systems like ultrasonic based and Lidar based systems fail to give accurate measurement in a laboratory test environment where workspace is small. With the developments in camera technology and availability of high speed processors, computer vision based measurement system is a viable solution in such indoor environments. In recent years, computer vision has played a major role in many sectors such as medical image processing, currency inspection, surveillance, movement recognition, object detection and tracking, etc. A computer vision system usually comprises of one or more camera for capturing the images, processing hardware system (computer), and software for processing the captured image. The required data from the captured images are extracted, processed, analyzed and given as an input to the controller which is carrying out the specified task. Most of the commercially available camera based detection and tracking system comes with a cost between 5-10 lack Indian Rupees making it uneconomical in many situations. Thus, in the current paper a low cost vision based measurement system is developed for analyzing the motion of a mobile robot in an indoor environment. Literature discusses and demonstrates many image processing techniques and algorithms for detection and tracking of static and mobile objects by utilizing their distinctive features depending upon the required application. Techniques such as modeling of image-background scene and foreground can be used for detection and tracking of mobile objects [1], [2]. In these techniques a model of the background scene is subtracted from the captured image to detect the location of object which is not included in the background model. Another approach for detecting a target object is with the help of kernel based technique which compares a set of models of the target object with the objects in the image. Kernel based tracking proved to give good performance in most of the image sequencing scenarios, though this approach involves complex computation [3]. Use of tunable kernels can be utilized to increase the accuracy of kernel based tracking. This method can efficiently track human motion [4]. Multicolored objects can be tracked by using Gaussian mixture model [5]. Boundary based and region based information of the target can be utilized for tracking deformable targets like biological cells [6]. Spatial features such as shape, contour, edges, and color can be exploited for serving the purpose of object detection and tracking in a given scenario. Among them, color acts as a powerful descriptor to identify and track the objects of interest in a captured scene. Tracking of desired object by using its color feature has been very effective since color conveys a variety of rich information that describes the object [7]. Color histogram based tracking can be utilized for various challenging situations such as tracking a football player, measuring position and orientation of the target object etc. [8]. Another application of color based tracking system is for tracking