Depth Perception Of Omnidirectional Structured Light For Noise Reduction

Information of range is an important field of computer vision research.At this stage,robots usually use range finder on the space for depth perception,but the high price was a key factor that hinder the range finder in promoting.The method of structured light range perception has high precision,but the noise interference is a key issue to be solved urgently,which serious restricts the development and application of structured light technology.Based on the principle of structured light depth measurement,an active omnidirectional depth measurement system is constructed based on the advantages of large field of view of the catadioptric panoramic camera combined with the ring structural light so as to obtain the depth information of 360 degrees in one shot.The main contents of this thesis are as follows:First,we design an omnidirectional depth measurement sensor and conduct the camera calibration research.A kind of omnidirectional depth measurement sensor based on structured light is designed by ourselves.Then the camera is calibrated according to the mathematical model of panoramic camera to obtain the intrinsic relationship between the main parameters of the device.Then,based on the moving robot collecting the structured light bar images under different noises environment in the indoor scene,the video data is divided into frames,and the sliding window method is used to intercept the region of interest for each frame image to make the dataset and analyze the noise in the image.Thirdly,based on the acquired structured light bar image datasets,convolutional de-noising auto-encoder network is used to preprocess the light bar images with different noise disturbances to reconstruct a clear background image with sharp outline and strong antiinterference characteristics.Finally,the center of the de-noised light bar is located based on the improved gray centerof-gravity method,and the depth distance of the space is calculated according to the fitting relationship between the pixel offset of the imaging object and the spatial distance.Through several sets of experimental data validation,the depth calculation result of the light bar after denoising is compared with the depth calculation results of the light bar denoised by the traditional image preprocessing method.In this thesis,the algorithm can get more dense depth point cloud,Improve the accuracy and robustness of depth measurement.At the same time,the efficient point cloud storage algorithm is designed to facilitate subsequent point cloud processing.