150results about How to "Meet the requirements of real-time processing" patented technology

The invention discloses a dual-aperture photoelectric imaging system-based unmanned aerial vehicle detection and recognition system. According to the dual-aperture photoelectric imaging system-based unmanned aerial vehicle detection and recognition system, a wide-angle camera and a long-focus camera are adopted as an unmanned aerial vehicle detection device and recognition device respectively for performing monitoring; and a monitoring image sequence is acquired, a target detection algorithm is adopted to detect an unmanned aerial vehicle suspected target, a pattern recognition algorithm is adopted to recognize the suspected target, and after the target is confirmed, the track is tracked, interfered and controlled. Photoelectric sensors are adopted as the unmanned aerial vehicle detection device and recognition device, and therefore, the dual-aperture photoelectric imaging system-based unmanned aerial vehicle detection and recognition system has the advantages of high reliability, low cost and the like; the suspected target is searched in a large space region by means of a wide-angle imaging system, and the target is confirmed and tracked by means of a long-focus imaging system which is a two-axis turntable, and therefore, requirements for high detection rate and high accuracy can be satisfied, the reliability of the system can be greatly improved, and therefore, the system can bring outstanding economic benefits and has high practical value.

The invention discloses an H.264 compression bit stream oriented video watermark method. The method comprises the following steps of: generating an authentication code according to a predictive mode of a 4*4 brightness block in an I film; and embedding the authentication code in the film-grade bit stream by modulating variable-length code words of some motion sub-blocks of a B film and a P film, wherein the modulation is performed based on a mapping rule between the variable-length code words and the bits to be embedded, and the mapped variable-length code words is extremely similar to the original variable-length code words. Because in the method, the video bit stream is unnecessarily pre-processed, and comparison and replacement operations are performed by selecting the proper variable-length code words only from the H.264/AVC compression bit stream in the watermark embedding process without a huge number of operations, so that the method has the advantages of low computing complexity, high safety and high practicability, and can better meet the requirements of the real-time video processing.

In the system, the input module is used for collecting the inputted digital image; the collected digital images is inputted into the image dissection module; the image dissection module dissects image into a light illumination image L and a reflection image R, both are respectively inputted into the light illumination processing module and the reflection processing module; the light illumination processing module makes non-linear correction for the light illumination image L and output an corrected light illumination image L'; the reflection module makes the partial noise reduction process for the reflection image R corresponding to the dark area of the inputted image and outputs a noise-reduced reflection image R'; the image combining and outputting module combines L' and R' and then displays the combined image on the output device.

The invention relates to a scanned document image error correction method. The method comprises the following steps: converting into a gray-scale image; performing filtering; extracting an edge-point binary map; by means of projection analysis, scanning an inclination angle of a scanned document image; and according to Q, calculating boundary of a text area, accordingly, calculating an offset amount, and by means of translation operation, enabling the text area to be centered. According to the invention, through detecting inclination of the input scanned document image, inclination of an original document image is rectified, and a document image with better visual quality can be obtained.

The invention relates to a method, a system and a device for synthesizing an image, belonging to the technical field of image processing. The method comprises the steps of: acquiring exposed images oftwo different exposure times of a target photographic subject, analyzing pixel point data of the same position in the two exposed images to obtain an image fused weight value, and fusing the two exposed images by using the weight value to obtain a composite image, wherein the composite image is synthesized by images of different exposure times, and the pixels at the same position are capable of capturing more image details, thereby improving the dynamic range of the composite image. At the same time, only two images with different exposure times are needed in the process of synthesizing the images, the amount of image calculation is reduced compared with the traditional technology, by which the synthesis process uses multiple images, which can meet the real-time requirements of image processing.

The invention discloses a parallel random sampling consensus (RANSAC) method. An assuming stage and a verification stage in the parallel RANSAC method are performed parallelly. Specifically, the method comprises the steps of parallelly acquiring m groups of random samples ai(i=1,2,...,m) from a data set A with N data numbers in the assuming stage and parallelly outputting m estimated models Mi(i=1,2,...,m) according to the m groups of random samples ai(i=1,2,...,m); adopting the m estimated models Mi(i=1,2,...,m) to parallelly vote the same point in the verification stage and gradually and directly judging whether each point in the data set A is an inside point or not. The invention further discloses a parallel RANSAC device. By adopting the parallel RANSAC method and the parallel RANSAC device, the performance and data throughout amount of an RANSAC algorithm can be improved, and the demand for high-speed real-time processing can be met.

The invention discloses a water falling detection and rescue control system based on deep learning, belonging to the field of computer vision, and the method comprises the steps: building an underwater personnel image data set; using a LabelImg tool to mark the position of a person in water and the category of the person in water in the original image, and storing the marking information generatedby each image in an xml file format for network training; using a YOLOv2 deep learning target detection framework i for training a data set, making clustering analysis on the established data set before network training, and dividing the data set into a training set, a verification set and a test set according to the proportion of 6: 2: 2. The rescue control system comprises a monitoring video input unit, a drowning person detection unit and an alarm rescue unit. The response efficiency of water surface rescue is effectively improved, and the survival probability of people falling into wateris improved.

The invention relates to a moving image processing device and a moving image processing method. The device and the method are based on an atmosphere model so that videos in foggy days are clear, the image visibility is improved, and in addition, the real-time processing requirements of the images can be preferably met. According to the moving image processing device, in the video defogging process, the video is divided into a core frame used as a main frame and an ordinary frame used as a sub frame, the core frame is calculated again to be used as a distance parameter t(X) and an unplugged point parameter A, the ordinary frame is not subjected to A calculation, and the A of the core frame is used, the t(X) of a region corresponding to the core frame is used for the background part of the ordinary frame, and the t(X) is calculated again for the foreground part of the ordinary frame. According to the device and the method, the application speed of the single-frame image defogging algorithm based on the atmosphere model in the defogging clarifying processing on the moving images such as videos and the like can be accelerated, a good defogging effect can be obtained, and in addition, the real-time performance of the moving images is ensured.

The invention discloses a non-adaptive airborne non-side-looking radar short-range clutter suppression method. A non-adaptive short-range clutter suppression method formed by space-time filtering and space-time matching is used for suppressing airborne radar short-range clutter. The method comprises the steps of firstly arranging received signals of all channels of radar into a column vector, then performing space-time filtering to the received signals to suppress clutter energy, and finally performing space-time two-dimensional matching to realize the detection of a target. The non-adaptive airborne non-side-looking radar short-range clutter suppression method overcomes the defects that a great number of training samples which satisfy independent and identical distribution conditions are needed when the traditional adaptive clutter suppression method is applied to airborne radar, the calculation amount is great and the performance during short-range clutter suppression is seriously decreased. Compared with the traditional adaptive clutter suppression method, the method provided by the invention has the advantages that since the method is a non-adaptive method, the method is not influenced by non-uniform samples, the calculation complexity is low, the clutter suppression can be conveniently and quickly realized and the detection of the target is more greatly facilitated.

The invention discloses a real time self-adapting processing method of anastigmatic imaging, which is a real time self-adapting processing method of anastigmatic imaging in a high-speed movement small-object imaging system. According to the method, small objects in high-speed movement can be automatically searched and found, FPGA and DSP are used as the real time on-line processing platform, an image reinforcement and inter-frame correlation identification algorism with low complexity is adopted, effect of fake objects are removed to obtain movement characteristics of the objects, so as to drive a high-precision two-dimensional piezoelectricity tilting mirror to complete self-adapting movement compensation, thereby realizing clear imaging of the high-speed movement small-objects.

The invention relates to the field of computer vision and machine learning, in particular to a method for detecting floating objects on water based on a YOLOv2 network. The method comprises the stepsof: 1, collecting data: 2, data enhancement: performing data enhancement of the data set A; 3, marking the picture: marking the floating object area on the water in the data set B with a rectangular frame, and 4, training the module: the purpose of dividing dataset B into three parts is to select the best weight model with the best generalization ability; 5, detecting the module: the trained weight model is used to detect the monitored river or lake video. The invention has the beneficial effect of replacing the traditional detection method based on artificiality, saving manpower and materialresources, and judging the pollution degree of a river or a lake. The invention randomly divides the data set into a training set, a test set and a verification set, and expands the sample through a data enhancement method to prevent the over-fitting problem caused by too few image samples.

The invention discloses a load identification method based on an improved probabilistic neural network. The load identification method comprises steps of adopting a binary system to perform coding on a load type in an electricity usage network, establishing a coding library of load types, collecting an electric parameter of each load type, establishing a non-linear mapping relation between the electric parameter and the code of each load type, creating a probabilistic neural network, training the probabilistic neural network to obtain an error function, using the error function as a fitness function of a particle swarm algorithm, adopting the particle swarm algorithm to perform optimization on a smooth factor of the probabilistic neural network to obtain an optimal smooth factor, updating the probabilistic neural network according to the obtained optimal smooth factor to obtain an improved probabilistic neural network and performing recognition on the load type in the electricity usage network on the basis of the probabilistic neural network. The load identification method based on the improved probabilistic neural network can simultaneously identify multiple loads, improves speed of identifying a pernicious load under a multi-load mode, and can better realize recognition of restricted electric appliances and electricity usage control under a condition of multiple electric appliances.

The invention discloses an abandoned object detection method. Brand-new environment self-adaptation double background models are built, and the background models are a front background model and a back background model respectively; independent update mechanisms are set for a first background image of the front background model and a second background image of the back background model; the first background image is updated according to a scene image; the second background image is updated according to the updated first background image; for a detected static target position, selective local updating is performed; the first and second background images are subjected to difference value operation; the influence of a shadow of the static target position in natural light on the detected static target position is eliminated by using a shadow inhibition technology; and the images are processed by a morphological processing algorithm, a contour extraction algorithm and the like to form an abandoned object region. The method has relatively high environmental adaptability, timeliness and robustness.

The invention relates to an IDL based method for three-dimensionally visualizing medical images. The method comprises the following steps of: converting DICOM data into three-dimensional visualized medical images through volume rendering algorithm, converting the medical images into a virtual reality data model through VRML interactive description, and controlling the displayed graphics state through a human-computer interface on a GUI (Graphical User Interface) and a control interface so as to form visualized virtual images, wherein the volume rendering algorithm comprises the following steps of: defining an object in an object graphics system; calling the object through an object-creating OBJ_NEW function; respectively calling a pattern object, a viewing object, viewing group objects and a scene object by utilizing ADD calling format; drawing the scene object in a window object by utilizing a window object DRAW method; and completely deleting the objects by utilizing the OBJ_DESTROY function after display is accomplished. The method has the advantages of meeting the basic requirements of visualization of the medical images, and being capable of carrying out corresponding operations, such as volume rendering, slicing, and the like.

The invention discloses a method for measuring a target azimuth angle by using a vehicle-mounted radar and a vehicle-mounted radar. According to the method and the vehicle-mounted radar, spatial domain echo signals are obtained according to the received echo signals of a target; an original spectral peak position corresponding to the peak value of an original spatial spectrum obtained according tothe spatial domain echo signals is determined; a to-be-estimated spatial spectrum angle range interval is determined based on the original spectral peak position; a to-be-estimated spatial spectrum angle is set; a preset APES algorithm is adopted for the angle so as to obtain the estimation result of a target spatial spectrum; and the target azimuth angle of the target is estimated based on the estimated spectral peak position of the target spatial spectrum. According to the method, the to-be-estimated spatial spectrum angle range interval is obtained through fast Fourier transform, so that the value range of the original to-be-estimated spatial spectrum angle can be narrowed; a Q matrix in the preset APES algorithm is obtained only according to a sample correlation matrix, so that the Qmatrix does not transform along with the to-be-estimated spatial spectrum angle; and inversion operation needs to be performed on the Q matrix once, and therefore, the calculation of super-resolutionangle measurement can be realized in a short time.