Research On Multi-object Recognition Based On Video

Video based multi object recognition is a computer vision,and it is a very important branch of the field of pattern recognition.When the control system controls the robot fish,the target tracking of the robot fish not only provides the exact position of the robot fish,but also needs to know the direction of the robot fish.These real-time data are the premise of robot fish's next step motion decision.Only by stable and accurate tracking of robot fish can the motion decision be carried out correctly,so as to control the robot fish correctly.This topic is the detection and tracking of robot fish,when the camera is fixed,the machine will be deformed fish swimming in the process,so the robot fish is on the tracking of non rigid object tracking,but light interference is serious,so we on the robot fish motion tracking of the main work is to pre select good modeling method and the tracking algorithm appropriate to late selection,stable tracking of robot fish.Moving target tracking algorithm is obtained by optimizing the current characteristics of the tracked object,to improve the accuracy of target search and matching,and effectively reduce the searching range by way of predicting the next target at the possible location of the target or to determine the search direction.In order to get a better perspective,this paper proposes a color space model based on HSV,we in the original color space re modeling,effectively remove the image in the main light and shadow interference,interference of the robotic fish of multiple robot fish image segmentation and recognition more accurate,so the improved algorithm for matching tracking algorithm was more robust.Based on the image segmentation of robot fish,this paper tests the CycleGAN network in the following chapters,and gets good results.