25 results about "Targeted detection" patented technology

The invention discloses a vehicle-mounted unit awakening method, device and system. The method comprises the following steps of obtaining at least one group of traffic flow images of a target area according to a target detection algorithm, wherein a traffic flow image is obtained through real-time shooting of a camera device corresponding to the target area, acquiring a traffic flow detection result located in the target area in real time according to the traffic flow image, and determining whether a vehicle-mounted unit located in the target area is awakened through a roadside unit corresponding to the target area or not according to the traffic flow detection result; when determining that the vehicle-mounted unit is awakened, sending an awakening signal to the vehicle-mounted unit through the roadside unit located in the detection area, so that the vehicle-mounted unit in the dormant state is awakened; when the vehicle-mounted unit is not awakened, indicating that the vehicle-mountedunit is in a dormant state, and when determining that the vehicle-mounted unit needs to be awakened according to actual road conditions, enabling the roadside unit to awaken the vehicle-mounted unit,so that the vehicle-mounted unit is prevented from being in an awakened state for a long time, the service life of the vehicle-mounted unit is prolonged, and energy consumption is reduced.

The invention discloses a video concentration tube set division method capable of reserving real interaction behaviors of a target, and relates to the technical field of image processing methods. The method comprises the following steps: extracting a target tube through target detection and tracking; dividing target tubes through a tube set optimization algorithm, wherein interactivity is measured by calculating space-time proximity and a direction interaction function; and generating a target management set according to the interaction measurement. According to the method, real interaction behaviors between targets in an original video can be reserved, and the effectiveness of the method is proved by experiments.

The invention belongs to the technical field of cylinder sleeve defect detection, and particularly relates to a cylinder sleeve surface defect detection method, system and device based on deep learning, and the method comprises the steps: collecting the real-time surface image data of a cylinder sleeve product on a production line in a regional manner, wherein the product surface image data acquired in different areas at least comprises images of the end face, the inner wall, the outer wall and the skirt edge of the cylinder; filtering and segmenting the image data to obtain to-be-detected area image data of the end face, the inner wall, the outer wall and the skirt edge of the product image; masking the image data of the to-be-detected area to determine the image data of the target detection area of the cylinder sleeve product; and using the trained and optimized classification network model to classify and identify the image data of the target detection area, and acquiring and removing cylinder sleeve products with defects on the surfaces by a sorting system. The invention is combined with deep learning technology to realize rapid and efficient detection of the surface defects of the cylinder sleeve, improves the efficiency, and has a good application value.

The invention belongs to the field of intelligent transportation, and relates to a fine-grained non-motor vehicle feature detection method, a storage medium and computer equipment, and the method comprises the steps: constructing a feature detection network which comprises a convolutional neural network, a plurality of block fusion modules and a driver and non-motor vehicle feature detector, wherein the block fusion modules which are connected are used for achieving a feature pyramid network from the deep feature map to the shallow feature map; selecting a plurality of network feature layers from the feature detection network, performing layer-by-layer blocking and fusion processing to obtain a driver fine-grained feature map and a non-motor vehicle fine-grained feature map, respectively inputting the driver fine-grained feature map and the non-motor vehicle fine-grained feature map into a driver feature detector and a non-motor vehicle feature detector for positioning and classifying regions of interest, and respectively outputting the regions of interest to an upper sampling layer; and further partitioning and fusing features to the shallow feature map to obtain a driver feature map and a non-motor vehicle feature map which are fused with finer-grained resolution information. According to the method, a detection network with block feature pyramid fusion is provided, and the target detection accuracy is improved.

The invention discloses a road-occupying operation monitoring method based on target detection and pedestrian tracking, and the method comprises the following steps: setting the position of a target monitoring region according to a video image obtained by a road monitoring camera, and carrying out the continuous image extraction of an input video stream according to a preset period, using the trained target detector to detect the lane occupying behavior; when the target detector detects that a road occupying behavior exists in the video image, starting to detect people appearing in the video image, and when people are detected, starting pedestrian tracking, and recording each person appearing in the video image and a movement track of each person; and analyzing the pedestrian trajectory data in the video stream, judging whether the pedestrian trajectory data accords with the road-occupying operation behavior characteristics or not, if so, triggering an alarm, and if not, ignoring the road-occupying behavior. According to the invention, by combining the moving track information of the person appearing in the monitoring scene, the road-occupying operation behaviors are automatically and accurately analyzed, researched and judged, temporary lane occupation and road-occupying operation events can be effectively distinguished, and the false alarm rate is reduced.

The invention discloses a moving target detection method and device, a storage medium and terminal equipment. The method comprises the following steps: acquiring a (t+1)th frame of image to be detected; judging whether each pixel point on the (t+1)th frame of image meets a first matching condition and a second matching condition or not according to a preset Gaussian mixture model, wherein the first matching condition is used for judging whether the deviation between the pixel value of each pixel point and the mean value of any corresponding background component is smaller than a first deviation threshold value or not, and the second matching condition is used for judging whether the number of the neighborhood pixel points of which the deviation between the pixel values and the mean value of any corresponding background component is smaller than a second deviation threshold value is greater than a number threshold value or not in the neighborhood pixel set of each pixel point; if the first matching condition and the second matching condition are not met, judging that the corresponding pixel point is a foreground; and obtaining a moving target area on the (t+1)th frame of image according to all the pixel points judged as the foreground. According to the invention, the field pixel information of the pixel points can be combined, and the detection effect of moving target detectionis improved.

The invention discloses a detection method and system based on a multi-sensor fusion drivable area, and a vehicle, and the method comprises the steps: carrying out the nearest distance solving and data association calculation of data points of each lane, carrying out the comparative analysis of an association result, taking the data point with the highest existence probability as a cut-off point of each lane, and finally outputting the cut-off point. By means of the method, the drivable area can be detected in real time, the maximum passable range of each lane can be continuously output, the defects caused by missing detection of a single target detection module are effectively overcome, the safety of the whole vehicle is improved, and an effective judgment basis is provided for planning and decision making of an unmanned driving system.

According to the technical scheme, the Tiny-YOLOv3-based pneumonia image target detection method comprises the following steps that (1) an original XML file is analyzed into TXT, and each row of the TXT file specifically comprises an image absolute path, coordinate values of the upper left corner and the lower right corner of a pneumonia focus and the probability that the focus exists; 2) the training data are trained by using a target detection model Tiny-YOLOv3, and finally a predicted value is obtained; (3) the difference between the position information and the probability information in the obtained predicted value and the position information and the probability information corresponding to the real value in the training set is processed through a loss function, and finally the convolution kernel parameter of the convolution operation in the feature extraction network is adjusted according to the processing result of the loss function; and 4) a to-be-detected pneumonia image is input to the trained target detection model, and the target detection model outputs a detection result which specifically comprises a coordinate value and a probability value. The method improves the model training precision, and enables the prediction result to be more accurate.

The invention discloses an image nonlinear interpolation acquisition method and system based on deep learning, and belongs to the technical field of image processing, and the method comprises the steps: 1, building a data set; 2, detecting an unqualified target in the interpolated image by using a deep learning target detection method to obtain a first target detection deep neural network; 3, detecting the position of an unqualified target in the original image and pixels near the unqualified target by using a deep learning target detection method, and obtaining a second target detection deepneural network; 4, based on the data set, the first target detection deep neural network and the second target detection deep neural network, performing training to obtain a deep neural network, and outputting a correct interpolation through the deep neural network. According to the invention, a deep learning target detection technology is used, a data set of artificial interpolation correction isconstructed, an image background is combined, an unqualified target and related nearby related pixels in a traditional interpolation algorithm are identified, and a correct interpolation is output incombination with an example of artificial interpolation.

The invention relates to the field of radar target detection, and provides a distributed radar target detection method and device based on a dynamic multi-scale grid. The method comprises the following steps: firstly, dividing a three-dimensional detection space by using a cuboid grid of which the size is far greater than the radar distance resolution, establishing an index between a large-size grid and each channel echo signal distance unit, and traversing a grid region to complete first target detection; and then iterative detection is carried out, that is, the detection range is continuously reduced according to the position of the existing target and the neighborhood range thereof obtained by last target detection, grids with smaller sizes are divided in real time in the reduced detection range, indexes of the grids with the smaller sizes and echo signal distance units of all channels are established, grid detection is carried out, and the detection accuracy is improved. According to the invention, the detection precision is ensured, the detection efficiency is effectively improved, the dynamic construction of the grid region and the index relationship is realized, and the occupation of the storage grid and the index relationship on memory resources is effectively reduced.

The invention relates to a shuttle vehicle automatic goods searching and positioning method and system based on deep learning, and the method comprises the steps: 1, collecting the images of various types of trays, marking the coordinates of the trays in the images, and making a tray training sample set; 2, inputting the tray training sample set into a built combined convolutional neural network model including a classification network and a target detection network, setting hyper-parameters, and then performing training; 3, saving the weights of the trained classification network model and the target detection network model; and 4, utilizing the classification network model and the target detection network model to classify and identify the images collected by the shuttle vehicle and control the rotating speed of a shuttle vehicle driving motor to accurately park and search goods. Automatic goods searching and positioning of the shuttle vehicle are realized by means of an image recognition target detection technology, the robustness of a shuttle vehicle positioning system is improved, the workload of field debugging personnel is reduced, and accurate positioning of the shuttle vehicle is ensured.

The field of this invention relates to immunohistochemistry (IHC) and in situ hybridization (ISH) for the targeted detection and mapping of biomolecules (e.g., proteins and miRNAs) in tissues or cells for example, for research use and for clinical use such by pathologists (e.g., biomarker analyses of a resected tumor or tumor biopsy). In particular, the use of mass spectrometric imaging (MSI) as a mode to detect and map the biomolecules in tissues or cells for example. More specifically, the field of this invention relates to photocleavable mass-tag reagents which are attached to probes such as antibodies and nucleic acids and used to achieve multiplex immunohistochemistry and in situ hybridization, with MSI as the mode of detection/readout. Probe types other than antibodies and nucleic acids are also covered in the field of invention, including but not limited to carbohydrate-binding proteins (e.g., lectins), receptors and ligands. Finally, the field of the invention also encompasses multi-omic MSI procedures, where MSI of photocleavable mass-tag probes is combined with other modes of MSI, such as direct label-free MSI of endogenous biomolecules from the biospecimen (e.g., tissue), whereby said biomolecules can be intact or digested (e.g., chemically digested or by enzyme).

The invention belongs to the technical field of target detection, and particularly relates to a lens module appearance detection method and system based on an SSD algorithm, and the method comprises the steps: 1, obtaining a sample image of a lens module, setting the feature information and category labels of the image, and obtaining a training set; 2, building an original SSD algorithm model; 3, optimizing the original SSD algorithm model to obtain a detection algorithm model; 4, training the detection algorithm model by using the training set; and step 5, using the trained detection algorithm model to carry out identification detection on the packaged lens module. According to the invention, on the basis of keeping the advantages of the SSD algorithm, automatic detection of the damage of the lens module can be realized.

The invention relates to an unmanned aerial vehicle line patrol image auxiliary collection method and system based on resolution reconstruction, and the method comprises the steps: carrying out the image collection through an airborne camera of an unmanned aerial vehicle, inputting the collected image into an airborne computing device, and carrying out the following steps: carrying out the preprocessing of the image, and carrying out the collection of the image; evaluating the definition of the input image through a definition evaluation algorithm, and if the definition of the input image does not meet a set threshold value, judging that the input image is a low-definition image and inputting the low-definition image into a resolution reconstruction model; reconstructing the low-definition image through the resolution reconstruction model, and converting the low-definition image into a high-definition image; inputting the input image of which the definition meets a set threshold value and the high-definition image output by the resolution reconstruction model into a target detection model for inspection target detection; the airborne computing device is arranged on the unmanned aerial vehicle, resolution reconstruction and target detection are performed on the acquired image, the inspection target is detected on line, and the image quality is improved.

The invention discloses a comprehensive detection method for an intelligent closestool. The method comprises the steps of obtaining first information which is model recognition information used for recognizing the type of the closestool at a feature recognition end, acquiring second information of the recognized closestool of the type according to the first information, wherein the second information has detection feature information of the recognized closestool of the type, determining the detection feature information of the closestool, and enabling the feature recognition end to move to a preset position away from the position of the to-be-detected feature according to the detection feature information, and obtaining the detection information of the closestool of the type, and uploading the detection information to a server to complete detection. Targeted detection can be carried out on different types of closestools, systematic diversified detection can be completed through a single detection device, and the detection process is accurate, efficient and stable.