Design Of High-precision Indoor Positioning System Based On Multi-source Information Fusion

The well-developed outdoor positioning services based on GPS and mobile base stations have been widely used in recent years.Meanwhile,with the advancement of 5G communication,relevant technologies and industries such as the Internet of Things,indoor robots and indoor navigation are transforming the positioning requirements from outdoors to indoors.However,since the well-developed outdoor positioning methods,restricted by such problems as signal occlusion and multipath reflection,fail to meet the needs of indoor positioning,further researches are required to find solutions.As the cost of image sensor and high-performance processor has been decreasing in recent years,the optical and vision-based positioning technology develops rapidly,and its positioning accuracy and output speed have exceeded other indoor positioning methods,such as WIFI,Bluetooth,ultrasound,RF and inertial navigation.Nevertheless,indoor visual positioning system cannot output position information very quickly due to the limitations from frame rate and traffic rate of camera.So far,the accuracy and speed of visual 3D positioning system still fails to meet the requirements of real-time control system(e.g.mobile robot)for indoor position control and trajectory tracking.Moreover,since the indoor environment is often susceptible to various interferences,the positioning system shall have excellent anti-interference performance and error detection capability.To meet the needs of indoor positioning,an indoor 3D positioning system based on visual orientation was set up.By using multiple cameras to observe a room in different directions,indoor space was covered with no dead angles.To reduce the influence of other light sources such as sunlight,lamp light and display on image object extraction,an algorithm based on the characteristics of the measured object was designed for accurate object extraction in accordance with the color and shape of the measured object.With the designed algorithm,the object's position in the coordinate system of pixels can be accurately and efficiently calculated by controlling the error within 0.5 pixels.An object classification algorithm based on Bayesian decision was also designed to accurately find the objects of the same color in images from different cameras by classifying the color of objects based on maximum a posteriori probability.Later,a new positioning method was designed to reconstruct the 3D position of object by using two or more camera images.The method has high positioning accuracy but no special requirements for optical axis of camera when it is installed.It can effectively solve the problem that the space orientation of camera cannot be accurately determined due to large span when the camera is installed indoors.Given that the positioning method merely based on vision may fail to the requirements of mobile robot for trajectory tracking because of low positioning speed,high-speed inertial sensors such as angular velocity sensor and acceleration sensor were introduced for multisource information fusion,so that the output speed of positioning data can be effectively increased while ensuring the positioning accuracy.In addition,complementary filter was used for frequency domain fusion of visual and inertial sensors since drift failure may occur in inertial sensors.In this way,nearly true measurements can be still obtained in the event of a failure.A more accurate true north can be also obtained in the absence of any compass device like magnetometer which is highly susceptible to interference in indoor environment.In this sense,the proposed method has a great potential in engineering application.The high-precision indoor positioning system above was designed to serve the project of dynamic evolution and formation control of multi-body system in the National Basic Research Program(973 Program).It can effectively satisfy the scenes with high requirements for positioning information,such as indoor positioning,trajectory tracking and non-cooperative target capturing of mobile robot.Moreover,it is able to simultaneously locate multiple objects and has certain resistance to a variety of common interferences indoors.The deviations that may occur during installation,calibration and reconstruction,together with the corresponding processing methods,were also described to provide a valuable reference for practical applications.