261 results about "Crowd counting" patented technology

This document discusses, among other things, methods and systems for determining the number of members in a group as well as changes over a period of time. Using an image of the scene, an overlap area is calculated by projecting portions of the image onto spaced apart and parallel planes. A filter correlates the overlap area to the number of members.

Embodiments of the invention disclose a crowd counting method based on deep learning and an apparatus thereof. The method comprises the following steps of dividing a crowd picture to be detected into a plurality of picture blocks; based on an RPN candidate frame acquired from training in advance, generating a model, and determining head candidate frame areas in the plurality of picture blocks and degrees of confidence of the head candidate frame areas; according to the degrees of confidence, screening the determined head candidate frame areas and acquiring an area to be detected; and based on a Fast-RCNN correction model acquired from the training in advance, carrying out classification prediction on the area to be detected, and according to a classification prediction result, determining a crowd number included in the crowd picture to be detected. In the embodiments of the invention, the crowd counting method under high resolution is provided, and accuracy and robustness of crowd counting are increased.

The invention discloses a crowd counting method and a system based on a scale pyramid neural network, wherein the scale pyramid neural network structure is composed of VGG-16 as the basic skeleton, and multi-scale high-level features are extracted by embedding a scale pyramid module in the high-level. The scale pyramid module consists of four parallel cavities with different expansion ratios. Themethod comprises steps: inputting a picture to network, after a series of convolution, getting high-level features and transfering them to scale pyramid module, generating four kinds of feature maps,these four kinds of feature maps and the feature maps input to the module being in the 'channel' dimension series connection, transfering to the next convolution layer, and finally getting the densitymap through the convolution of 1*1. Compared with the prior art, the method of the invention has more efficient, more accurate and simpler network structure, and can apply modules for handling multi-scale problems to deep networks.

The invention provides a dense crowd counting method and a device thereof. By obtaining an image to be detected including a portrait, the image to be detected is input into a convolution neural network model to obtain a crowd density map of the image to be detected, and the number of portraits in the image to be detected is determined according to the crowd density map. The process fully extractsthe feature information from the image to be detected, realizes good crowd counting and density estimation, and brings great convenience for subsequent safety monitoring, crowd control and other applications.

The invention discloses a crowd counting method based on a coding-decoding structure multi-scale convolutional neural network, and the method is characterized in that the method comprises the following steps: considering the visual angle distortion of an image, and employing an adaptive Gaussian filter to calculate a true value density map of the image; adopting an encoding-decoding structure, building a multi-scale convolutional neural network model, wherein a loss function of the multi-scale convolutional neural network model comprises pixel space loss and counting error description; training and testing the built multi-scale convolutional neural network model to obtain a trained multi-scale convolutional neural network model; inputting a to-be-estimated image into the trained multi-scale convolutional neural network model, and predicting to obtain a crowd density map; and performing regression estimation on the crowd density map to obtain the number of people in the to-be-estimatedimage. According to the method, scale features and contextual information of the image can be reserved, and the output quality of the density map can be improved.

The invention discloses a dense crowd counting and personnel distribution estimating method based on deep learning, which includes the following steps: preprocessing a target crowd image to be detected; selecting an appropriate data set to train a designed deep learning network model; inputting target data to the trained network for analysis and identification; and finally, getting a thermodynamic diagram of people in the coverage area, and getting the number and distribution of people in the crowd. Through the method, the number and distribution of people in a high-density crowd can be estimated accurately, and the problem that people in a crowd are blocked due to high density is well solved. The method is applicable to video images under any lighting condition, in any environment background and of any resolution.

The invention discloses a crowd counting method based on a convolutional neural network. The method comprises the following steps of (1) after a training picture is marked, performing convolution operation with a Gaussian kernel to obtain a real crowd density diagram, and taking the real crowd density diagram as a label for model training; (2) inputting the training picture and the corresponding real crowd density diagram into a convolutional neural network model for performing training, and optimizing iterative updating parameters every time until the model is converged; (3) creating a new scene data set, and finely adjusting the obtained model by utilizing a model migration method, thereby finishing the model training; and (4) performing performance evaluation and testing on the trainedmodel. By utilizing the method, the number of parameters needed to be trained for the model is reduced; the model structure is simplified; the real-time performance of the model is improved on the premise that the accuracy is guaranteed; and the requirements of actual application are met.

The invention discloses a static image crowd counting method based on joint learning, includes the following steps: pre-training stage: using 50-layer residual network to train the ImageNet 2012 classification data set to obtain the parameter initialization classifier network, classifying the image blocks into three categories by a Softmax, respectively corresponding to three regressors; In the training phase, every image block of the training data set is input to three regressors respectively, and the same image block will get different counting results. The regressor with the smallest counting error is used as a classification tag to mark the image block, and the three kinds of image blocks are used to fine-tune the regressors respectively. Classifier training phase: random sampling andensure that the number of labels of each category is consistent; Joint training phase: continuous iterative training of classifiers and regressors. The invention can carry out counting and density estimation in crowd dense scene, has certain scale self-adaptability, and improves counting precision and model generalization ability.

The invention discloses a crowd counting method based on deep learning target detection. The crowd counting method comprises the following steps of constructing a deep learning network model: adoptinga YOLO3 network taking a DarkNet network as a basic network; processing the training data, obtaining crowd image data under multiple scenes, and expanding the scale of the training set in an image mirroring processing and scale random interception manner; and training the network: optimizing the network through the loss function and the gradient descent training parameters. The invention aims toovercome the defects of the existing crowd counting. In a specific environment, a target detection method based on a deep neural network is adopted to count and count crowds, so that the problem of low accuracy in a traditional feature extraction method is solved, and also the problem that errors are large under the condition of crowd sparsity in a deep learning feature regression method is solved, the detection speed is greatly increased, the detection speed is four times the speed based on a 101-layer residual network RetineaNet (retinal network), and the precision is equivalent to that of the 101-layer residual network RetineaNet (retinal network).

The invention discloses a dense crowd people number counting method and system based on a deep convolutional neural network. The counting method comprises the following steps that: obtaining an original image which contains a dense crowd, dividing the obtained original image into a plurality of image blocks with the consistent size, and recording an original image label to which each image blocks belongs; utilizing the image block and the original image label to which each image block belongs to train the deep convolutional neural network; and utilizing the deep convolutional neural network which finishes training to calculate the people number in each image block, carrying out accumulation on the people numbers in all image blocks, and finally, obtaining all people numbers in the original image. By use of the method, crowd counting accuracy and robustness can be effectively improved.

The invention discloses a crowd counting method based on scene depth information, which comprises the following steps of: (10) image segmentation: extracting depth information of an input image by adopting a monocular image depth estimation algorithm, and segmenting the input image into a close-range region and a distant-range region according to the depth information; (20) close-range area crowdcounting: predicting a head boundary frame of the close-range area by adopting a crowd counting method based on target detection to obtain the number of people in the close-range area; (30) crowd counting in a distant view area: predicting a crowd density map in the distant view area by adopting a crowd counting method based on density estimation, and solving the product of the crowd density map in the distant view area to obtain the number of people in the distant view area; and (40) confirmation of the number of people in the input image: adding the number of people in the distant view areaand the number of people in the close view area to obtain the total number of people in the input image. The crowd counting method based on the scene depth information is good in multi-scale adaptability, low in overfitting degree and accurate in counting.

The invention provides a crowd counting method based on a multi-scale context enhancement network, which comprises the following steps: inputting a picture, obtaining shallow features and deep features after feature extraction, performing feature fusion through a feature fusion module, sending the fused features into a multi-scale perception module, and finally, encoding the space and channel information of the features through a context enhancement module to obtain a density map with crowd distribution features. The estimated number of people in the current picture can be obtained by summingthe density image pixels. According to the crowd counting method based on the multi-scale context enhancement network, the multi-scale problem existing in crowd counting can be effectively solved, andmore accurate counting and density estimation can be carried out on crowds in a complex scene by modeling space and channel context information of a feature map. The invention is high in robustness,and can provide accurate data for the safety and planning of a large crowd gathering place.

The invention belongs to the technical field of computer vision and intelligent transportation, in particular to a generalized density crowd counting method based on multi-scale depth learning, whichis called PaDNet. The invention uses a plurality of sub-networks to learn specific crowd scale characteristics. Firstly, the data set is clustered, and the data set is divided into a plurality of density grades: low density data set, medium density data set and high density data set. Using the data of a specific density level to pre-train a specific sub-network, and then a scale enhancement network is used to enhance the scale characteristics; Finally, all the scale features are merged into the final density map through a converged network to count them. The invention can enable the specific sub-network to learn the accurate scale characteristics, so that different networks can identify the crowd characteristics of different density grades and more accurately count the crowd.

Robust crowd counting is an important yet challenging task of interest in a number of applications and is compounded by the fact that crowd behavior is unpredictable. Traditional approaches that tackle these issues are mainly classified into two categories: video based recognition and non-image based localization. However, both suffer inherent drawbacks such as being limited to line-of-sight or requiring people to carry specific electronic devices which limit them even within relatively well controlled environments without considering public locations or rapidly emerging events. Accordingly, it would be beneficial to provide a system that implements a device free crowd counting methodology with no impact on network capacity and/or performance.

The invention discloses a method for counting persons in a video based on LSTM (Long Short-Term Memory)-weighted neural network, and the method comprises the steps: firstly estimating a perspective according to different scenarios, and then generating an adaptive density map of the crowd; carrying out the down-sampling of continuous multi-frame images, inputting the images to a neural network, andtraining the network based on the images and its corresponding real adaptive density map; estimating the density map of the input images through the trained network, and predicting the number of persons based on the density map. According to the scale difference of the crowd in the scenarios, the method measures the different scale features obtained through network learning through a scale loss function. The method employs a weighted loss function for balancing the contributions of different regions for solving a problem that the crowd is not uniform in distribution. Meanwhile, the method obtains the correlation information of the adjacent frames through LSTM, performs the postprocessing of the predicted number of persons through smooth filtering, and improves the person counting accuracy.

The invention relates to a crowd counting model construction method, counting method and device, and the method comprises the steps: carrying out the head labeling of a crowd image in a pre-stored data set, obtaining a crowd density map, and enabling the crowd image and the crowd density map corresponding to the crowd image to be combined, and obtaining a target data set; performing data amplification processing on a training set obtained by dividing the target data set according to a first preset proportion to obtain a target training set; and finally, training the multi-scale feature aggregation convolutional neural network model after parameter initialization processing by using the training crowd image and the training crowd density map in the target training set to obtain a crowd counting model. By the adoption of the technical scheme, when the model is trained and applied, the problem of scale change of figures in the image can be solved, the calculation amount can be reduced, the recognition efficiency can be improved, a perspective drawing does not need to be used in the training process, and the applicability of the model is improved.

The invention discloses a people counting method based on video analysis. The method comprises the following steps of inputting a video image, acquiring a foreground picture by using a background subtraction method, and clustering the foreground picture into a plurality of blocks; extracting features, and performing perspective correction on the features; estimating the number of people by using a two-layer regression model, wherein the first-layer regression divides the different blocks in each frame into different disperse density layers, and the second regression combines virtual standardization and regression counting as a joint learning process; training a counting model synchronously covering virtual standardization and regression problems for the different disperse density layers; and at last counting each block by using the different regression models according to the different crowd density layers, and accumulating the numbers of all blocks to obtain the number of people. According to the people counting method, based on the two-layer regression model, and through combining the vision normalization with the number regression, the defect of a single regression model is overcome, and better robustness and adaptability are provided for shielding multi-density crowd scenes and crowds and segmenting incomplete images.

The invention discloses a crowd counting and future pedestrian volume prediction method based on video images. The method comprises the steps: 1, selecting a video image data set with annotation information, conducting Gaussian function processing according to annotation of a head position, and generating a true value density map; 2, inputting a video frame into a built MPDC model to extract a feature map, and mapping the feature map into a crowd estimation density map (DE); and 3, inputting obtained DE stacking frames into a constructed Bi-ConvLSTM network, predicting a crowd prediction density map at a T+1 moment, and estimating the number of pedestrians at the T+1 moment. According to the method, a convolutional network based on a multi-scale pyramid cavity and a Bi-ConvLSTM network based on residual connection are adopted, a crowd estimation density map is generated by using continuous video frames, a crowd prediction density map of a future frame is further predicted, and the number of crowds is counted. The method aims at the prediction of continuous video images, is a brand-new method, can obtain a real-time crowd density map and the number of people, and also can predict the crowd density map and the pedestrian volume of a future frame.

The invention discloses a multi-scale dense crowd counting method, and belongs to the field of aviation monitoring. Firstly, collecting data of a dense scene, marking and preprocessing crowd images toserve as training pictures, and then sequentially subjecting the training pictures to convolution operation and a multi-level pooling module to obtain feature maps corresponding to the pictures and fusing with multi-scale information, and using a convolutional layer with a convolution kernel of 1 * 1 step length of 1 to respectively perform positioning information enhancement on each feature mapto obtain the corresponding feature map with enhanced positioning information, and repeatedly using the convolution operation and the multi-level pooling module to fuse each positioning information enhanced feature map, re-positioning the information enhancement to obtain a final feature map, decoding, and gradually recovering the spatial resolution by using a bilinear interpolation method to obtain respective final crowd density map, and carrying out integral summation by using the pixel value in each crowd density map to obtain the final number of people. According to the invention, the counting precision is improved, and a better cognitive ability is provided for a monitoring scene.

The embodiment of the invention provides a crowd counting method, a network, a system and an electronic device. The method comprises the steps of obtaining a to-be-detected image; inputting the to-be-detected image into a neural network, and carrying out feature extraction; and processing the features extracted by the neural network, and counting the number of people. By utilizing the method disclosed by the invention, a high-quality crowd density map can be outputted in the crowd counting problem, and meanwhile, the high-accuracy crowd counting estimation can be obtained.

The invention discloses a density crowd counting method based on the HOG characteristic and the color histogram characteristic. The method includes the following steps that S1, the HOG characteristic of an acquired image region is extracted, and the color histogram characteristic of a detection window in which the HOG characteristic is generated is extracted; S2, a first SVM classifier is acquired based on HOG characteristic training, and a second SVM classifier is acquired based on color histogram characteristic training, wherein the first SVM classifier is used for pre-estimating a detection region, and the second SVM classifier is used for secondarily classifying the pre-estimated detection region; whether the detection region is a head region or not is determined by weighting and combining results acquired by the two SVM classifiers; S3, according to the detected head region, the number of crowds in a video can be counted through a region matching method based on an optical flow method. The density crowd counting method based on the HOG characteristic and the color histogram characteristic is good in accuracy and anti-interference performance.

The invention discloses an intelligent campus anti-treading early warning device based on video crowd counting statistics technology. The device is characterized in that the device comprises a first person statistical camera installed on the entrance of the campus stairs; a second person statistical camera installed at the exit of the campus stairs; an early warning server located in the campus monitoring center; the first person statistical camera and the second person statistical camera are respectively connected to the early warning server; the early warning server is further connected witha campus broadcast horn, an evacuation guide indicator light and an warning device; the early warning server is connected to the receiving module through a wireless communication module. The device adopts a personnel statistical camera to perform real-time monitoring on the main entrances and exits of the campus. By counting the number of personnel, and analyzing the flow of personnel, and predicting the behavior of personnel through video analysis algorithms, early intervention reduces the risk of stampede events that occur on campus students in tight spaces, improves emergency evacuation, rescues efficiency, and improves campus safety management.

The invention provides a crowd counting method and system, a computer-readable storage medium and a server. The crowd counting method comprises the following steps: training and preprocessing an original image set containing a crowd to form a preprocessed image set; combining with multi-channel deformable convolution, constructing a convolution neural network, wherein the convolution neural network is used for mapping an original image set containing a crowd to a thermodynamic map; training the convolutional neural network to form another convolutional neural network.; inputting a test set ofthe original image set into the other convolution neural network to obtain a thermodynamic map corresponding to the original image set. The crowd counting method and system, the computer readable storage medium and the server provided by the invention can effectively solve the crowd counting problem under a plurality of scales and a plurality of scenes.