35results about How to "To achieve the effect of real-time detection" patented technology

The invention discloses a method for real-time detection of a road vehicle based on a deep learning SSD frame and mainly solves a problem of low detection accuracy of the existing technology under the conditions of traffic congestion and complicated weather. An implementation scheme of the method comprises the following steps that 1.shooting videos of a plurality of running vehicles at a vital communication line and grabbing window information of the vehicle in each frame of image in the video in a manual labelling way; 2.taking VGG-16 in a classification network as a basic network, constructing an SSD300*300 detection frame, extracting features of some feature extraction layers to carry out connection as recognized feature vectors and inputting the feature vectors into a loss function; 3.training through a training sample and taking a trained detection model into the detection frame; and 4.setting a correlation threshold and detecting the vehicles in the test videos by the trained detection model. According to the method, the detection accuracy is greatly improved and an effect of real-time detection is realized, and the method can be used for vehicle detection under the complex scene.

The invention provides a ship detection method and a ship detection system combining significance detection and depth learning, comprising constructing a ship image sample library, comprising collecting coastal area surveillance video data under visible light, obtaining a frame image containing the ship from the video data, and then obtaining a true value of the ship position, length and width; constructing a YOLO-like convolution neural network to train the video ship target samples and obtaining the training result model of the ship target in the surveillance video; Inputting the ship imagedata and predicting the detection boundary box by using the training result model; by using boundary box information, detecting the saliency to obtain more accurate ship position. The technical schemeof the invention is very fast and efficient, and can achieve the effect of real-time detection. For complex scenes such as clouds and fog, cloudy weather, rain and other conditions, the method can also have good detection results, and has high robustness.

The invention disclose a digital electronic shear speckle interferometer based on a micro-polaroid array. The digital electronic shear speckle interferometer comprises a laser, a beam expander located on a light path of an outgoing beam of the laser, a to-be-detected object, a shearer located on a light path of a beam reflected from the surface of the to-be-detected object and used for obtaining two mutually perpendicular coherent linearly polarized lights, the micro-polaroid array arranged on a light path of a beam outgoing from the shearer, a photosensitive element connected with the micro-polaroid array, and processing equipment connected with the photosensitive element and used for transmitting and processing a signal acquired by the photosensitive element. The interferometer has the advantages that multiple phase shifts can be obtained by single exposure; a real-time detection effect can be achieved; a phase shift device is simplified in the light path; and as the micro-polaroid array generating the phase shifts is directly integrated onto the photosensitive element, the antivibration requirement is lowered.

The invention discloses an unmanned driving platform real-time target detection system based on a camera and a laser radar. Space-time synchronous pixel-level fusion is performed on original data of acamera and a laser radar; a clustering detection result is obtained by combining the obtained space-time synchronization data with a data analysis method of the laser radar; an improved faster-rcnn network architecture is constructed to perform parameter training and real-time detection; and the type, length, width and height of a target object around an unmanned driving platform, the distance ofa central point relative to the space coordinate of the unmanned driving platform, a yaw angle, a roll angle and a pitch angle are output. According to the system and a detection method thereof, a traditional clustering and artificial intelligence fusion algorithm is adopted, the defects that the image detection space positioning precision is low and the category is difficult to judge through point cloud detection are overcome, and a real-time 3D target detection system based on a camera and a laser radar in an unmanned driving platform scene is realized.

The invention discloses a method for detecting abnormal behaviors in an elevator car based on computer vision, and the method comprises the steps: obtaining the external contour of a moving target through combining a ViBe algorithm of an inter-frame difference method, and obtaining the height-width ratio information of the body of a passenger in the car on the basis of the external contour; and detecting the number and the positions of the heads of the human bodies in the lift car through the YOLOv3 neural network. According to the number of people detected by the YOLOv3 neural network, the abnormal behavior detection in the car is divided into a single-person situation, a two-person situation and an above two-person situation. Under the condition of a single person, the falling behavior is detected by utilizing the change of the height-width ratio of the human body contour of the passenger and the vertical movement distance of the head of the passenger. On the basis of a target contour obtained by combining a ViBe algorithm of an inter-frame difference method in a scene of two or more persons, the average kinetic energy of passengers is calculated by utilizing a pyramid LK opticalflow algorithm, and the detection of violent infringement behaviors is realized. Video input and detection are carried out at the same time, and the effect of real-time detection is achieved.

The invention discloses a high-voltage wire defect detection method based on a regional convolutional neural network. The method comprises the steps of obtaining a high-voltage wire image defect sample set; inputting the high-voltage wire image into a convolutional neural network, and extracting defect features in the high-voltage wire image to obtain a feature map; inputting the feature map intoan area proposal network additionally provided with a full connection layer, generating a proposal area by the area proposal network, and judging the proposal area; inputting the proposal area and thefeature map which belong to the foreground into a full connection layer for defect classification; correcting and processing the boundary of the proposal area, and optimizing the boundary to obtain amore accurate proposal area; and repeating the steps to iterate the regional proposal network until the training is completed, and obtaining the high-voltage wire defect detection model with the trained parameters to carry out actual defect detection. According to the invention, the Faster R-CNN algorithm is used to identify and position the wire defects, and the method has the advantages of highidentification speed, extremely high accuracy and good robustness.

The invention discloses an intelligent monitoring method, device and system based on machine vision, equipment and a storage medium. The method comprises the following steps: obtaining a monitoring video in a production environment, extracting a plurality of selected frames in the monitoring video, and dividing dangerous areas in the selected frames into a plurality of sub-areas; calculating the fingerprint of each sub-area by using a Hash algorithm, and calculating the Hamming distance between each fingerprint and the center of a plurality of clustering clusters, wherein the plurality of clustering clusters are obtained by performing fingerprint clustering on each sub-area in the video frame; and according to the relationship between the Hamming distance and the maximum distance in each clustering cluster, judging whether abnormity occurs or not. The intelligent monitoring method based on machine vision provided by the invention is simple and effective, does not need a large amount oftraining data, can achieve the purpose of real-time detection, and can meet the safety requirement of real-time shutdown due to the detection accuracy.

The invention is a real-time detection device for surface defects of additively manufactured products, which includes a lighting module connected to a computer processing unit, a lens group and a sensor module, a transmission module and a computer processing unit; the lighting module works when the lens group shoots process; the lens group includes one or more optical lenses; the transmission module includes a stepping motor, a screw and a corresponding control circuit, which is used to change the height of the lens group so that the lens group can always follow the manufacturing of the additive product Process; the computer processing unit is used to receive the images collected by the lens group, and after merging the collected images, judge and record the surface defect type and position of the additive product through the corresponding image processing algorithm. The invention can acquire 360-degree surface images of additive products at the same time, detect product surface depressions, protrusions, discoloration and other surface defects comprehensively and without dead angles, perform real-time detection with high precision and high resolution, and improve the detection speed of the system.

The invention provides a personnel detection method and a device suitable for big data, equipment and a medium, and the method comprises the steps: S1, taking one frame from each continuous M frames in an image sequence of a video, and obtaining a sampling video sequence; S2, extracting continuous N frames from the sampled video sequence each time, and inputting the N frames into a pre-classifierto obtain a preliminary to-be-detected human head window; S3, inputting the preliminary to-be-detected human head window into the main classifier, and further excluding the human head-free window to obtain a secondary to-be-detected human head window; and S4, inputting the secondary to-be-detected human head window into the clustering analyzer, mutually correcting the detection results of each frame, reducing errors, and generating a high-precision human head detection result. According to the method, head detection in various scenes with many people is realized based on the multi-stage cascaded head classifier, static features, dynamic features and inter-frame correction are fused, and the technical problem that good head detection effect is difficult to obtain due to interference of complex backgrounds and insufficient detection precision in actual scenes is solved.

The invention discloses a vehicle-mounted intelligent traffic light recognizing and prompting device. The vehicle-mounted intelligent traffic light recognizing and prompting device comprises an equipment layer; a data layer is arranged at the connecting end of the equipment layer; the equipment layer comprises a control unit; a traffic light detection unit, a voice broadcasting unit and a data storage unit are arranged at the connecting end of the control unit; the data layer comprises a background server; a WeChat official account unit and a cloud memory are arranged at the connecting end ofthe background server; and a user mobile phone is arranged at the connecting end of the WeChat official account unit. According to the vehicle-mounted intelligent traffic light recognizing and prompting device, the color of a traffic light is identified by using the traffic light detection unit, the traffic light detection unit transmits recognition information to the control unit after recognition, the control unit reads the recognition information and converts the recognition information into a corresponding instruction, the corresponding instruction is broadcast by using the voice broadcasting unit, the broadcasting mode is a mode of broadcasting in sequence, such as left turn, forward only and no right turn; and therefore, a real-time detection effect is realized, the broadcasting accuracy is guaranteed, a vehicle owner is prevented from being misguided, the design is reasonable, and the popularization is made convenient.

The invention discloses a method for detecting the weaving density of a three-dimensional woven fabric based on deep learning. The method comprises the following steps: dividing a three-dimensional fabric surface data set into a training set and a verification set, and constructing a deep learning network model for training and verification; the deep learning network model can accurately detect the classification and position information of each longitudinal and transverse line monomer on the surface image of the three-dimensional fabric, then outputs a detection result based on the model, and calculates the density of the longitudinal and transverse lines by using a statistical method according to the coordinates of a prediction frame for longitudinal and transverse line detection. The method for detecting the knitting density of the three-dimensional knitted fabric is high in accuracy and efficiency, can achieve a real-time detection effect, reduces human subjective interference, improves the quality of the three-dimensional knitted fabric, and guarantees the reliability of product production. Meanwhile, the blank of the automatic detection technology for the weaving density of the three-dimensional woven fabric is filled, and the efficiency of detecting the consistency of the three-dimensional woven fabric is improved.

The invention relates to the technical field of medical apparatus and instruments, in particular to a medical vehicle automatic disinfection system. According to the system, the technical design thatan integrated circuit AD654 and a programmable logic device GAL cooperate to generate a pulse signal is adopted as core technical design, so that the system controls a stepping motor through a single-chip microcomputer to realize stepless speed regulation, and then the effect of uniformly spraying disinfectants is achieved; by adopting the mode that a power tube drives a motor to work and the design method of motor overcurrent protection, the effects that a disinfection vehicle can arrive at a designated position requiring control at any moment, the disinfection vehicle is parked at an arbitrary angle, and the motor is under overcurrent protection are achieved; by adopting the mode that a relay in a switch module controls liquid extraction time and the design method of liquid level detection, the effects of electrical control of liquid preparation and automatic detection of liquid remaining conditions are achieved; and by adopting the technical method that a wireless module is used asa communication bridge to remotely receive information, the effects of real-time monitoring of motor running states and remote control of a medical vehicle are achieved.

The invention provides a detection card for an industrial robot drive board. It includes a detection card, a single-chip microcomputer, a multi-channel analog switch, a golden finger and a test signal buffer; wherein, the single-chip microcomputer, a multi-channel analog switch and a test signal buffer are inserted into a pre-configured chip socket on the detection card; the The multi-channel analog switch and the test signal buffer are electrically connected to the single-chip electromechanical device; the multi-channel analog switch is electrically connected to the test signal buffer; the detection card is inserted into the preset card slot of the tested drive board through the golden finger; Wherein, when the multi-channel analog switch is turned on, the single-chip microcomputer sends a PWM motor driving signal to the checked driving board, and receives a feedback signal from the checked driving board to determine whether the checked driving board is faulty.

The invention provides a method, a system and equipment for detecting violation setting of an electric power tower operation fence, and the method comprises the steps: constructing an image database, marking an obtained to-be-detected image, and storing the marked to-be-detected image into the image database; taking a worker on a tower as a detection target, training a construction scene detection model, and forming a scene detection model; a fence on the ground is used as a detection target, a fence detection model is trained, and the fence detection model is formed; taking the ground in the picture as a segmentation target, training a ground segmentation model, and forming the ground segmentation model; and processing a video stream collected at a construction site into pictures, and respectively inputting the pictures into the scene detection model, the fence detection model and the ground segmentation model to judge whether an illegal behavior exists or not. According to the invention, the workload of manual inspection operation in a construction site is reduced, the occurrence of leak detection is prevented, and the occurrence of safety accidents caused by illegal operation of constructors is avoided.