Abstract
There has been exponential growth in the use of computer vision to solve real world problems in advanced systems like autonomous cars and faster auto-focus in smartphone cameras. There is a constant state of improvement in this field as more novel techniques are being developed to solve various problems. This project is an attempt to improve the speed estimation technique in computer vision systems. The speed of an object can be estimated using optical flow and several different algorithms to help concentrate on the object. Optical or optic flow is the pattern of apparent motion of objects, surfaces, and edges in a visual scene caused by the relative motion between an observer (an eye or a camera) and the scene. This is an established technique to estimate the speed of an object in the scene. This system is replicated using the OpenCV (Open Computer Vision) libraries where we use a camera connected to a computer as input. Accuracy of the estimation can widely vary if the distance of the object from the camera is unknown. Visually, the RGB video is also important as the difference in color of the object and the background can make it easy for the filters to isolate the object of interest from the background, provided that we use an effective background subtraction technique. Special camera arrays with two cameras or also known as stereo camera can process depth information similar to human vision system. Two sets of images are used to calculate the depth of the objects in the image using a disparity map. This information can play a vital role in various cases. This project uses the Microsoft Kinect sensor which has a built in depth sensor and a RGB camera to obtain the depth within its field of view. The depth data from the Kinect sensor exposes us to a wealth of new information. We use this depth data and the plane data from the camera to estimate the speed of the object. After this system is built, we performed several set of controlled tests and estimate the speed of the object and the fidelity of the system including the fidelity of the depth sensor. Using this map, we then find the speed of the object in real time.