Research Of Indoor Target Search For A Mobile Robot With The Hybrid Vision System

The visual target search in the robot communities has been a popular problem in resent years.It's of great importance for the intelligent robot to effectively search target and obtain the target position in the scene.In previous works,many target searches utilizes monocular camera or binocular camera to acquire the scene information.This kind of pictures acquired by the traditional camera is clear and has high resolution,but the field of view is limited.In order to obtain the global scene characteristic in scene reconstruction task,where multiple viewpoints must be selected to acquire a map of the whole scene.This increases the system complexity and reduces the efficiency of information detection.An omni-directional camera with a 360° horizontal field of view,can contain more information about the environment.Then,with the development of the RGB-D camera,it can obtain the depth information in RGB-D image.A hybrid vision system,which includes a RGB-D camera and a hyperbolic mirror omni-directional camera,has been built.In this dissertation,a method of indoor target search for a mobile robot with the hybrid vision system was proposed.Several key problems such as detecting significant areas,positioning,path-planning and significant areas relocation were analyzed and discussed.The main results include:(1)Based on RGB-D camera and omni-directional camera system,a hybrid vision system was built.The internal and external parameters of the system were calibrated and the coordinates of the two camera system were unified.Then,the method to find the common view area between two cameras was presented.According to the depth information,the 3D reconstruction of the SURF points in the common view area was implemented.(2)The traditional visual attention mechanism is complicate and the detection accuracy is low.In order to overcome these defects,a new fast visual salience object detection method was proposed,which was based on fusion depth information on indoor RGB image.It can quickly access to significant number of objects,as well as the location of the relationship between each other.(3)In order to revise the regional initial position error and deviation of the robot movement,an improved Mean-shift algorithm based on motion changes was presented.Beside on this algorithm,the robot was relocated during the moving process.Finally,the object search in significant areas has been tested.