General Target Tracking Based On Information Merging Of Machine Vision And Machine Hearing

In indoor environments,target tracking is becoming more and more important.Target tracking is significance for robots to construct the indoor maps,track the moving object and create 3D scene.At the same time,worldwide researchers are also more and more aware of the importance of tracking persons for the future development of machine vision.The traditional tracking methods are based on 2D data.Because of the limitation of the two-dimensional image,a series of tracking failure are present,especially when objects are occluded and objects move fast.It becomes an important problem to solve for intelligent robot.In the cluttered environment,researchers want to search out the target object to take a long time tracking with the disturbance the other similar objects.It becomes a common concern how to extract and fuse the audio data and visual information with the mature technology.For handling the limitation of the twodimensional tracking methods,this paper studies the audio recognition and localization methods as well as proposes a novel 3D tracking method based on sound information.For audio information recognition and localization,this paper mainly studies a localization method based on time-delay estimation and analyses its performance.We also propose and demonstrate a novel random forest network based on the binary characteristics.We discuss a variety of weighted generalized cross correlation time delay estimation methods in the performance and improve the cross power spectrum phase(CSP)weighted generalized cross correlation time delay estimation(GCC)method.Compared with before improvements,the simulation results show that the new GCC-MCSP method has perfect anti-noise and antireverberation and that the positioning precision is improved.In addition,we also discusses localization methods based on time delay,the key is the geometric positioning method based on time delay estimation in azimuth,and sum up the relations between the topological structure and performance of the microphone array,whose error was analyzed.For image processing,we analyze the importance and feasibility of algorithm improvement.When the objects move in high-speed or is partly occlude,the system will lose the target occurrence.Tis situation is related to algorithm design.According to these drawbacks about this algorithm,we discuss how to adjust the track fusion algorithm to achieve more robust tracking result.For information fusion,this paper presents a new sound information assistance fusion filtering algorithm,which ingeniously applies sound information transfer matrix.It replaces the traditional image pyramid to reduce the computation time and memory.At the same time,due to the addition of sound information,system can track the fast target speaker.Even some speakers are occluded by others,they still can get better effect under dark condition.The system has generalized performance,which has great application value.Based on the above theories,the mobile robot based on Kinect is tested through the whole system and each modules...