52results about How to "Target detection is accurate" patented technology

The invention relates to an aided driving system based on a collision early-warning algorithm, and belongs to the technical field of computer vision and intelligent aided driving. The system comprises a detection and distance measurement module which collects road condition information in the driving process of an automobile through a camera, and carries out the detection, recognition and distance measurement of an obstacle through a YOLOv3 model; a collision early-warning module which is used for carrying out the collision prediction classification, calculating the time required by collision, giving early-warning judgment in time and carrying out early-warning broadcast on a driver; a positioning module which is used for acquiring driving position information of the vehicle by utilizing GPS/IMU integrated navigation, automatically switching the system to an IMU for positioning when a GPS signal is lost, and switching the system to GPS positioning again when the GPS signal is normal; and a GUI display and cloud video backup module which is used for displaying the identification video stream, the driving state and the map software annotation information in real time and carrying out cloud backup. According to the invention, the prediction precision and real-time performance of the auxiliary driving system can be improved.

The invention discloses a depth convolution network optical remote sensing image target detection method based on extended residual convolution, which solves the problems of low detection accuracy rate and high false alarm rate of optical remote sensing image plane and ship in the prior art. The implementation steps are as follows: constructing a test data set; construct training data set; constructing a target detection network based on extended residual convolution for extended feature receptive field,training target detection network based on extended residual convolution using training data set; using the trained target detection network based on extended residual convolution to detect the target from the test data set, outputting test results. The network constructed by the inventionis more suitable for target detection of an optical remote sensing image by using an expanded residual convolution module and feature fusion, and not only improves the accuracy of a common target, butalso obviously improves the accuracy of small target detection for the optical remote sensing image. The method is used for object detection in optical remote sensing images.

The invention discloses a laser radar rotation control method for target detection. The laser radar rotation control method comprises the steps that S1, a laser radar is installed on an operating platform to form a detection system, the change trends acquired at different operating speeds of the operating platform that point cloud distributions change with the scanning speed of the laser radar areobtained; S2, when the detection system performs target detection, the operating speed of the operating platform is acquired in real time, the change trend of corresponding point distribution is obtained according to the operating speed acquired in real time, and the current required laser radar scanning speed is determined according to the change trend of the point distribution; S3, the laser radar is controlled to rotate according to the scanning speed of the laser radar determined in the step S2 so as to complete target detection. The laser radar rotation control method has the advantagesof being simple to achieve, small in working blind zone and large in effective detection distance and making scanned point cloud data rich.

The invention discloses a target detection method based on global convolution and local deep convolution fusion, changes an original three-dimensional area suggestion network, and provides an ASD network structure based on asymmetric segmentation depth perception for target detection. By doing so, the features of each level and depth in the feature map can be extracted more fully. In addition, innovative technologies such as horizontal and vertical convolution fusion networks, a distillation network and an angle optimization algorithm are introduced, so that the detection effect is further improved.

A multicopter includes: motors to respectively rotate three or more rotors; and a radar system to transmit and receive a signal wave and detect a target by using the signal wave. An object detection apparatus in the radar system transmits and receives a signal wave to perform a target detecting process. An antenna element is in a position to receive the transmission wave reflected off a rotor (a rotor-originated reflected wave). The signal wave received at the antenna element is inclusive of a target-originated reflected wave reflected off a target and a rotor-originated reflected wave. The apparatus determines whether or not a frequency band satisfying a predefined condition for identifying a frequency peak is contained in a frequency spectrum of the signal wave as received by the antenna element, and determines a peak of a frequency band satisfying the predefined condition to be a frequency of the target-originated reflected wave.

The invention belongs to the technical field of multi-source data hierarchical fusion and extraction based on a multi-source sensor, and particularly relates to a hierarchical fusion and extraction method for moving target multi-source detection. The method comprises the following steps: performing registration fusion on a visible light image and an infrared light image to obtain a first-layer fusion image; registering the first-layer fusion image and the hyperspectral image, and weakening the registered image pixels according to the ground object classification area to obtain a second-layer fusion image; and performing target detection on the second-layer fusion image to obtain position information of the target in the image, sensing the target to obtain longitude and latitude of the target in a real environment, and adjusting an attitude of the aircraft to track the target to realize continuous detection and sensing of the target. In combination of multiple image sources signal features of the multiple image sources are effectively combined through image fusion, redundant repeated data information is removed, the accuracy of target detection is improved, and the detection efficiency is improved.

An electronic scan type radar device which uses a high resolution performance processing to estimate directions of arrival of radio waves, wherein powers of arrival waves received for targets are accurately calculated, that is, a vehicle-mounted radar device utilizing electronic scan which uses a predetermined angle estimation system to estimate directions of arrival of reflected waves, comprising finding mode vectors for angles calculated from the receive signals of the antennas, decomposing a vector of the receive signals into directions of the mode vectors, and defining the lengths of the decomposed vectors the receive powers of the reflected waves arriving from the targets. Due to this method, even if there are targets, it is possible to accurately calculate the powers of the arrival waves, whereby pairing is accurately performed, the precision of detection of targets is improved, and erroneous operation of a vehicle-mounted radar device utilizing electronic scans is prevented.

The invention discloses a device and method for detecting a high-speed tiny target online in real time by simulating fly vision. The method comprises the following steps of: acquiring scene video information by using a binocular camera, transmitting the scene video information into a DSP (digital signal processor) chip, and performing primary vision processing; performing large scene and small scene integration and target detection on primary motional information by using an FPGA (field programmable gate array) chip; and tracking a tiny target moving at high speed by taking an integration result of a large scene and a small scene as a target detection evidence. The device and method disclosed by the invention have the advantages that: the target detection is realized by virtue of a biological principle; the device and method have relatively strong antijamming capability and are applicable to the target detection under the condition of a low signal-to-noise ratio in various severe natural environments; a neuron integrating mechanism of a fly vision system has the characteristics of simple computing principle, high real-time performance and the like; the fly vision neuron is simple in tissue structure and can be easily realized by hardware; and the device can be installed on an automobile, panzer, airplane and other appliances which move at high speed, and has the capability of accurately detecting the high-speed tiny target on line in real time in a dynamically changing background.

The invention provides a kindergarten robot morning check system which comprises a body temperature detection device, a weight detection device, a height detection device and a herpes detection system. The herpes detection system comprises a workstation, a cloud server and a robot client, wherein the robot client acquires user image data and transmits the user image data to the cloud server, the cloud server receives the image data of the robot client and distributes the image data to the workstation, the workstation conducts recognition and detection on herpes, the palms and the oral cavity with a target detection and recognition algorithm based on a deep learning neural network and feeds a recognition result back to the cloud server, and the recognition result is transmitted to the robotclient for display through the cloud server. The kindergarten robot morning check system creatively utilizes the cloud server and a deep learning neural network model to recognize herpes, the operation is simple, the detection speed is high, and the detection result is accurate.

The invention belongs to the field of underwater target detection, and specifically relates to a method for labeling a sea cucumber target detection result by using a rotatable boundary frame. The method comprises the following steps of: performing data expansion on a sea cucumber training data set manufactured by using labeme software; constructing a full convolutional neural network; Carrying out offline training on the constructed full convolutional neural network by utilizing the expanded data set; Inputting the image containing the sea cucumbers into the trained full convolutional neuralnetwork to obtain a segmented image; Performing corrosion and filtering operations on the segmented image to obtain a post-processing segmented image; And searching a maximum connected domain on the post-processing segmentation map, namely the detected sea cucumber target. According to the method, the obtained segmented image is corroded, burrs on the periphery of the sea cucumber are removed, itis guaranteed that the minimum circumscribed rectangle is more accurate, the sea cucumber grabbing pose cannot appear outside the sea cucumber, and positioning is more accurate.

The invention discloses an unsupervised oil tank target detection method based on a shape-guided significance model. The method comprises the steps of: inputting a remote-sensing image, calculating anedge response graph of the remote-sensing image, and carrying out clustering on all pixels in the remote-sensing image to form superpixels to obtain all the superpixels of the remote-sensing image; obtaining a plurality of clustering regions on the basis of all the superpixels and the edge response graph; utilizing the clustering regions to obtain round probability and a round probability graph;calculating a shape guidance-based significance graph according to all the superpixels and the round probability graph; obtaining a binary result graph through shape-guided significance graph; utilizing a binary result graph to mark oil tank regions in the remote-sensing image to obtain the target regions. Through the target detection method, oil tank targets in low-resolution remote-sensing images under different sizes and illumination conditions can be accurately detected, and the method has better robustness.

A target detection apparatus acquires an image by imaging the surroundings of a vehicle, detects a self-position of the vehicle based on an amount of movement of the vehicle from an initial position thereof, and estimates a relative position of a target, located around the vehicle, with respect to the vehicle based on the self-position and information on a position of the target on a map. The target detection apparatus sets a detection area for the target within the image based on the relative position of the target with respect to the vehicle, and detects the target from the detection area. The target detection apparatus estimates error contained in the self-position based on the amount of movement of the vehicle from the initial position, and adjusts the size of the detection area for the target according to the error.

The invention discloses a target detection method. The target detection method comprises: the following steps: using a first semi-supervised learning model trained by labeling data to perform image detection on an area image acquired by a vehicle-mounted camera, and acquiring image detection data; training the first semi-supervised learning model on the basis of the image detection data and pointcloud data acquired by a laser radar, and obtaining a second semi-supervised learning model for target detection based on the second semi-supervised learning model. The invention also discloses a target detection device and a computer-readable storage medium. The invention can make up the deficiency of target observation by human eyes and improve the driving safety.

The invention belongs to the field of computer vision, particularly relates to an incremental small sample target detection method and system based on weight generation, and aims to solve the problems that an existing target detector lacks the capacity of small sample rapid learning and incremental learning, is high in dependency on label data and does not have openness. The method comprises the following steps: performing detector supervision training through basic category data; obtaining weights of scale perception and centrality perception of the basic category target detector, and generating a basic category response; generating a new category weight in combination with the basic category response; performing fine tuning training of the basic category target detector in combination with the new category data; and realizing incremental small sample target detection through the obtained target detectors of the basic category and the new category. According to the method, scale and centrality perception is combined, regional features are more representative, target positioning is more accurate, the model can obtain better overall performance in incremental learning, and detection efficiency, accuracy and precision are high.

The invention discloses an unmanned aerial vehicle aerial image target detection method based on improved YOLO V5, and belongs to the field of deep learning and target detection. The method comprises the following steps: constructing a related data set by using aerial images of an unmanned aerial vehicle; secondly, replacing a slice layer in a Focus module by using a convolutional layer in a YOLO V5 backbone network part; using the Neck part to further process the image features; then, for the problems of target stray distribution and too small target pixel ratio caused by a high-altitude aerial photography view angle of the unmanned aerial vehicle, optimizing and eliminating a 76 * 76 * 255 large detection head in a network prediction layer part, and adjusting an anchor frame at the same time; and finally, evaluating target detection performance through generalization intersection-union ratio, average precision and reasoning speed. According to the method, on the basis of improving the recognition accuracy and the feature extraction performance, the unmanned aerial vehicle aerial image target can be rapidly and accurately detected.

The invention discloses a multi-view road intelligent identification method based on dual-light fusion. The method comprises the following steps: collecting a visible light image and an infrared image of a road; preprocessing and fusing the visible light image and the infrared image to obtain a fused image; performing three-dimensional reconstruction on the visible light image and the infrared image to obtain a three-dimensional point cloud image; and training and detecting a pre-constructed deep learning network according to the fused image and the three-dimensional point cloud image to obtain a road recognition result; forming a depth map through image fusion and multi-view synchronous photography so as to detect the accurate size, position and distance of a target. When the system is applied to an automatic driving scene, the vehicle can detect any object under any road, light and weather conditions, and the all-weather application scene of the automatic driving system is greatly expanded.

The invention relates to a weak supervision target detection method based on image attribute representation learning, and belongs to the technical field of image processing. The method sequentially comprises the steps of label text description data processing, image feature extraction and target suggestion box extraction, label text feature construction, text and image feature fusion-based pseudo ground-truth mining, image attribute learning and prediction module, target classification and target suggestion box regression. According to the method, the interpretability of target classification is improved through image attribute learning, the mined pseudo group-truth is more accurate by utilizing text feature and image feature fusion, and the detection capability of the weak supervision model is improved.

An embodiment of the invention provides a target detection method for unmanned driving, equipment and a storage medium. The target detection method comprises the steps of: collecting an original imagein an environment where an unmanned vehicle is located, carrying out feature extraction on the original image, and generating a feature map tensor; performing convolution operation on the feature maptensor by using a plurality of convolution layers, generating a plurality of first target feature map tensors in sequence, and performing deconvolution operation on the first target feature map tensors corresponding to the last convolution operation to generate a plurality of deconvolution feature map tensors, wherein the deconvolution feature map tensors are in one-to-one correspondence with thefirst target feature map tensors, and the sizes of feature maps in the deconvolution feature map tensors are equal to the sizes of feature maps in the first target feature map tensors; and generatinga target detection result according to the deconvolution feature map tensors and the first target feature map tensors. Therefore, the target detection precision is improved, the precise detection ofthe unmanned vehicle on the target object is realized, and the vehicle driving safety is improved.

The invention relates to a one-stage direction remote sensing image target detection method based on student-T distribution assistance, and solves the problem of frames in any direction by using a geometric method based on a horizontal frame. The method comprises the following steps: S1, converting a remote sensing image by using a geometric conversion method based on the horizontal frame; S2, extracting remote sensing image features; S3, carrying out the regression and classification on feature maps obtained from a Convolutional Neural Network (CNN) and a Feature Pyramid Network (FPN) respectively, and extracting the feature maps of the feature maps of the CNN and the FPN from the feature maps of the Feature Pyramid Network; S4, carrying out the result optimization and output: adopting student-T distribution as a result of joint distribution, synthesizing a classification branch and a regression branch, optimizing the one-stage direction detection model based on the student-T distribution, and outputting a target detection result. According to the student-T distribution assistance-based one-stage direction remote sensing image target detection method, the special rigidity and bird's-eye view characteristics of the remote sensing image target are fully utilized, and the CNN and FPN models based on deep learning and Gaussian distribution and inverse gamma distribution are adopted, so that more accurate remote sensing image target detection is realized.

The invention provides a target detection method and device, computer equipment and a storage medium, and the method comprises the steps: firstly obtaining a to-be-processed image, carrying out the target detection of the to-be-processed image, and obtaining the depth information and two-dimensional block diagram information of each target object in the to-be-processed image; obtaining three-dimensional center information, dimensionality and orientation information of each target object according to the depth information and the two-dimensional block diagram information of each target object in the to-be-processed image, and finally determining three-dimensional block diagram information of each target object according to the three-dimensional center information, the dimensionality and the orientation information of each target object. In the technical scheme, adaptive voxel processing is performed on the depth information and the two-dimensional block diagram information of each target object, and scaling processing is performed on the two-dimensional block diagram information of each target object, so that target detection is performed on the to-be-processed image more accurately, and the problem of low detection accuracy in the prior art is avoided.

The invention belongs to the field, particularly relates to a video target detection method, system and device based on class external memory, and aims to solve the problem that the target detection performance is obviously reduced when the number of auxiliary frames is small in the prior art. The method comprises: firstly, training a video target detection model through a self-attention mechanismaccording to training video information; and obtaining enhanced instance features of the to-be-detected video through the trained video target detection model and the self-attention mechanism, and finally inputting the enhanced instance features into a classification branch and a bounding box regression branch of the universal target detection network to obtain a target detection result. According to the invention, the sensitivity of a video target detection method based on feature integration to the number of auxiliary frames in the prior art is reduced, so that target detection can be accurately carried out under the condition of fewer auxiliary frames or no auxiliary frames.