187 results about "Non uniform illumination" patented technology

The invention relates to technology for measuring electromagnetic scattering properties of targets, and aims to provide an effective and feasible symmetric polarization RCS (radar cross-section) testing method for targets at near fields. An observation antenna rotationally scans along a peripheral track, the position and the posture of the observation antenna are changed, simulative symmetric polarization non-uniform illumination fields are combined, multiple times of target scattering measurements are successively carried out and realize coherent combination, symmetric polarization RCS measurement for the targets at the near fields is realized, and the shortcoming that measurement of polarization electromagnetic scattering data of the targets at the near fields is incomplete is overcome.

The invention discloses a reading method of a pointer type meter based on depth learning, belonging to the field of depth learning and computer vision. The method of the invention utilizes Mask- RCNNobject detection and instance segmentation algorithm to divide the dial and pointer images firstly, then correcting the dial by perspective transformation, Then using PCA (Principal Component Analysis) algorithm to fit the segmented instrument pointer, Then judging the direction of the pointer according to the center coordinate of the smallest oblique circumscribed rectangle of the pointer and thecenter coordinate of the dial; Finally, calculating the pointer reading according to the slope and direction of the pointer by using the angle method. As that method of the invention can accurately classify the type of the instrument, high precision pixel level segmentation of the pointer and dial, under non-uniform illumination conditions, different scales still have good robustness, to solve the traditional pointer instrument recognition field of the dial and pointer positioning difficulties, uneven light, mirror reflection, blurred pictures caused by low recognition accuracy.

The invention discloses an underwater image enhancement method in combination with bilateral filtering and Retinex. The method comprises the steps of converting an RGB color space of an input image toan HSV color space; for a brightness component, filtering image noises in combination with the bilateral filtering, and separating out a reflective component and an irradiation component based on Gaussian center surrounding multi-scale Retinex for the denoised brightness component; for the separated irradiation component, adjusting illumination by adopting gamma correction to obtain a new irradiation component; realizing fusion enhancement of the irradiation component and the reflective component based on a Retinex theory, thereby obtaining a brightness component after non-uniform illumination correction; recombining the enhanced brightness component with a hue component and a saturation component, and converting the HSV color space to the RGB color space; calculating grayscale value ratios of red and green channels to a blue channel, and remapping the grayscale values of the three channels; and finally performing histogram equalization on the R, G and B channels of the image to realize image contrast enhancement. According to the method, the blue and green color deviation of the underwater image is effectively eliminated, so that the image contrast is improved; and the partial image noises are filtered, so that the signal-noise ratio of the image is increased.

The present invention discloses a panorama stitching brightness equalization method and system. The method comprises: by acquiring multiple paths of input images used for stitching a panorama, calculating a brightness mean value and a brightness difference value, which respectively correspond to overlap regions of each path of input image and two adjacent paths of input images stitched to both sides of the input image; according to the brightness mean value and the brightness difference value, which respectively correspond to the overlap regions at both the sides of each path of input image, and respectively corresponding positions of the overlap regions at both the sides on each path of input image, calculating brightness correction values that respectively correspond to both the sides of each path of input image; and according to the brightness correction values respectively corresponding to both the sides of each path of input image and a position corresponding to each pixel point on each path of input image, carrying out correction on a brightness value of each pixel point on each path of input image. According to the method and the system, each path of input image can be stitched and fused into the panorama with excellent overall brightness and natural transition under the non-uniform illumination condition; and the brightness integrity and consistency of the panorama are improved, and great convenience is brought to users.

An optical system for forming an image of at least a portion of an illuminated area on an object may comprise a lens positioned a spaced distance from the illuminated area on the object and an aperture stop positioned so that it is substantially co-planar with the image side focal plane of the lens. An occluding element is positioned between the lens and the illuminated area on the object so that the occluding element blocks a predetermined amount of light from a brightly illuminated region in the illuminated area but does not substantially block light from a less brightly illuminated region in the illuminated area.

The invention relates to a method for identifying a license plate in a high-definition image, which comprises steps of candidate license plate area detection, grayscale image binarization, obliquity correction, character segmentation, character identification and post-processing. Compared with the prior art, the method for identifying the license plate in the high-definition image has the advantages of simultaneous and quick positioning and identification of a plurality of license plates in a complex background with high definition (more than 5,000,000 pixels), good image binarization effect under the condition of non-uniform illumination, variability of license plate obliquity and size in a relatively larger range, and the like, and simultaneously greatly reduces the requirements on the installation of a video camera.

The invention discloses a packaging and code spraying detection method based on deep learning. The method comprises character zone extraction, single-line character segmentation and single-character identification. The method comprises the following steps that: using a semantic segmentation network to carry out character zone extraction on an image; then, using a column classification network to segment single-line characters; and finally, using a zone classification network to identify the single character. By use of the method, each network is mutually independent and is free from coupling,and finally, final identification accuracy is high. The interference of external bad conditions, including noise, distortion, low contrast, non-uniform illumination and the like, can be reduced to a maximum degree, and the character zone is accurately extracted and can be correctly segmented and identified. Most importantly, the method realizes a purpose that high detection accuracy can be achieved by small sample training, and in addition, the method achieves the requirements of industrial circles on the aspects of detection time consumption and accuracy.

The invention relates to a go identification method based on a go image and a go board. The go identification method based on the go image comprises the following steps that cross-shaped and T-shaped feature points are identified by searching the go board image; the four edge lines of the go are found according to the T-shaped feature points; then the positions of the four angular points of the go board are obtained; 17 ideal transverse and vertical go board lines and 361 go board line cross points are calculated through the positions of the four angular points of the go board; the cross-shaped feature points are found near the ideal go board lines and corrected; and go pieces are identified at the positions of the 361 go board line cross points so that go manual data are obtained and the outcome is calculated. The go identification method based on the go image is effective in processing method and high in reliability and can be adaptive to the conditions of a certain degree of distortion and non-uniform illumination so that identification accuracy can be enhanced.

Techniques for modifying data of an image that can be implemented in a digital camera, video image capturing device and other optical systems are provided to correct for non-uniform illumination appearing in data obtained using one or more illumination sources from a two-dimensional photo-sensor. In order to correct for these variations, a small amount of modification data is stored in a small memory within the optical system. According to a specific embodiment, non-uniform illumination correction factors are derived during a calibration procedure by illuminating a surface having uniform optical properties with the non-uniform illumination source and imaging that illuminated surface onto the photo-sensor.

Embodiments of the invention relate to decoding encoded barcode information from a series of camera-captured images by generating binarized sequences from the series of camera-captured images and decoding the encoded barcode information from the binarized sequences. A series of preprocessed images is generated from the series of camera-captured images by removing non-uniform illumination, reducing noise, and enhancing contrast between bars and spaces of the series of camera-captured images. Barcode-image orientations are estimated for the series of preprocessed images. The series of preprocessed images are projected along their respective barcode-image orientations to generate a series of gray-sequence images that represent barcode information in one dimension. The binarized sequences, each of which contains partial barcode information, are connected into a concatenated sequence according to computed shift information of neighboring binarized sequences.

In a variable shaped lamp shade of an LED lamp, the lamp shade is made of a translucent material matched with an LED lamp strip and a lamp holder and includes at least one strip-shaped optical refraction unit having an external refractive surface, an internal refractive surface corresponding to the external refractive surface, and an assembling structure for matching the lamp holder. The external refractive surface or internal refractive surface is a curved surface without an inflection point and the curved surface has a constant or gradually changing curvature; and a non-curved surface is formed on the other side. The variable shaped lamp overcomes the problems of conventional LED lamp strips having a low illumination and a non-uniform illumination caused by a direct projection or an installation of a conventional lamp shade, and a low light utility caused by a too-large illumination range.

The invention discloses a multi-target unmarked attitude estimation method based on a deep convolutional neural network. The multi-target unmarked attitude estimation method comprises the following steps: acquiring an image sequence containing a target behavior from a video; for each image, marking the position and category of each target feature part, and constructing a training data set and a test data set; constructing a multi-target attitude estimation model; training the multi-target attitude estimation model to adjust the weight of the multi-target attitude estimation model, and optimizing the multi-target attitude estimation model; and carrying out precision evaluation on the multi-target attitude estimation model, carrying out target attitude estimation or retraining according to arelation between precision and an expected value, and optimizing the multi-target attitude estimation model. According to the multi-target unmarked attitude estimation method, the accuracy can reachthe human level only through a small number of training images; the target feature part can be effectively tracked under the conditions of messy background, non-uniform illumination, camera distortionand the like; and a good effect can be achieved for different types of targets.

The invention provides an underwater colored image restoration method based on color correction and a red channel priori. A Retinex algorithm is used to eliminate non-uniform underwater illumination; a Lambert diffuse reflection model is combined with a gray world algorithm to estimate the color of a light source, light is introduced to improve an underwater attenuation imaging model, and the color of the light source is eliminated to realize color correction; and a prior method of a red color is used to restore a color corrected underwater image. Thus, the method provided by the invention can be used to restore the underwater colored image effectively, and eliminate non-uniform illumination and color distortion.

The invention relates to the field of intelligent traffic, in particular to a lane line adaptive detection method. According to the method, an image vanishing point attribute principle is utilized, the coordinate transformation is completed, and an interesting region is determined; the interesting region is subjected to layering and blocking, and an edge image is subjected to binarization processing; the line segment features are counted, and the descrambling processing is carried out according to the width features and the brightness features of lane lines; and a table of a projection straight line required in Hough conversion is built, the distance feature of left and right lanes and the mutual parallel features of the two lane lines are combined, and the properest straight line is found out to be used as the current lane line. The lane line adaptive detection method has the beneficial effects that through the layering and the blocking on the images, the algorithm is processed in a local small region; then, the dividing processing is carried out on the edge image, the influence of non-uniform illumination is reduced, and the detection effect on the lane lines with low quality is enhanced; the interference due to characters, vehicles, shadow and the like can be effectively avoided, and the detection accuracy is enhanced; and cos(Theta)*w and sin(Theta)*h are realized in a table look-up mode, and the calculation quantity of Hough conversion is effectively reduced.

A color enhancement method for use in a video camera imaging system (10,30) in the context of non-uniform lighting includes an adaptive algorithm (100–110) for developing a shutter speed control signal, analog gain control signal and color balance signals from pixel luminance over one field of video. The control signals developed during one field are used to control the camera (12) during the next field. The speed control signal and analog signal are determined based on the luminance signal obtained in digital form from the video camera, and analyzed over at least one field of video. During one field, the speed control signal and the analog gain signal target dark areas of the image and during the next field they target the bright areas of the image so that, over two fields, every pixel is within a workable range of luminance in at least one of two alternate fields. As the color balance settings for the fields are adjusted according to the exposure settings and the luminance level, consistent hue values are maintained between alternate fields. Additionally, a pixel color detection circuit (18,28) examines each pixel from the video camera (12) to determine whether it is in a color region of interest. The pixel color detection circuit uses a color offset signal, calculated by a controller (16,60) based on the luminance signal during the preceding field. The color offset signal indicates how far from neutral, or white, in the chosen color space, is the chrominance data from the camera. The video camera and pixel color detection circuit may be used to identify weeds or plant material on the ground. A controller (74) is responsive to the pixel color detection circuit and activates one or more spray nozzles ((68,70) when a weed is detected.

The invention discloses an image enhancement method in a dark light environment and belongs to the field of image processing. The problems of distorted representation result, blurred image contour andlow color saturation after enhancement seriously exist can be solved; The method comprises the following steps: extracting a brightness component V in an HSV in a dark light image, carrying out logarithmic transformation of global brightness, and carrying out overall brightness improvement on a low-illumination area in the dark light image; Obtaining neighborhood average brightness of pixel points in the current dark light image by using bilateral filtering under a receptive field three-Gaussian model of the retinal neurons; A function of extracting a local bright part from an image background is realized by utilizing top and bottom cap transformation processing, and then light and shade extraction and noise reduction are carried out on the image by utilizing top and bottom cap transformation; According to the method, the enhanced details of the dark-light color image can be well represented, the problem of non-uniform illumination distribution is solved, the overall contrast of the image is improved, the detail contour of the image is clear, and the color saturation is high.

The invention discloses an iris locating method and apparatus. The method comprises the steps of obtaining image information of a target iris, and performing rough locating on a pupil in the image information; according to a rough locating result of the pupil, constructing a tripartite graph of a target object, and according to the tripartite graph, extracting the target object from the image information through a preset algorithm, wherein the target object comprises the pupil, an eyelid and an iris excircle of the target iris; and according to the extracted pupil, eyelid and iris excircle, determining locating information of the target iris, wherein the locating information comprises at least one of the following information: a pupil boundary, an eyelid boundary and an iris outer boundary. The technical problem of inaccurate iris locating result when an iris image is a non-ideal iris image with the defects of non-circular pupil, non-uniform illumination, over-exposure or under-exposure and the like in the prior art is solved.

The invention discloses a maximum power point tracking method for a photovoltaic power generation system under non-uniform illumination. The method comprises the following steps of: detecting the output voltage U[k] and current I[k] of a photovoltaic array at the current moment, judging whether non-uniform illumination exists in the photovoltaic array through a maximum power point tracker, performing a variable step size perturbation and observation method to obtain the maximum power point if the illumination is uniform, and calculating the first reference voltage V0 if the illumination is non-uniform; obtaining the first local maximum power point by the variable step size perturbation and observation method, and recording corresponding voltage V1 and power P1; obtaining the second local maximum power point by the variable step size perturbation and observation method, and recording corresponding voltage V2 and power P2; by parity of reasoning, obtaining the voltage Vn and Vp, comparing n local maximum power points, extracting the maximum local maximum power point and taking the corresponding voltage as reference voltage Vnf+ output by the maximum power point tracker; and finally, calculating to obtain a duty ratio through a controller according to the Vnf+, and tracking the local maximum power point. The maximum power point tracking method has the advantages that the power loss brought by non-uniform illumination in the photovoltaic array in the system is reduced.

The invention relates to a hand area segmentation method deeply integrating significance detection and prior knowledge. This method combines the significant pattern of a hand area at pixel level with the significant pattern of a hand area at the regional level to make the detection algorithm for the entire hand area achieve higher robustness and accuracy. The method comprises the following steps: using an introduced Bayesian framework to obtain the confidence degree of each pixel in the hand area; combining the relevant techniques such as threshold segmentation, and finally obtaining a hand area segmentation result with high accuracy. The invention overcomes the shortcomings of a traditional hand area method which can only be applied to relatively simple background and non-near-skin color interference scene, and the method can obtain a hand area segmentation image accurately even in the case of various disturbances such as non-uniform illumination, near skin color background and face noise, therefore, making it applied with great prospects.