CN108399362B - Rapid pedestrian detection method and device - Google Patents

Rapid pedestrian detection method and device Download PDF

Info

Publication number
CN108399362B
CN108399362B CN201810069322.XA CN201810069322A CN108399362B CN 108399362 B CN108399362 B CN 108399362B CN 201810069322 A CN201810069322 A CN 201810069322A CN 108399362 B CN108399362 B CN 108399362B
Authority
CN
China
Prior art keywords
layer
network
target
neural network
convolutional neural
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810069322.XA
Other languages
Chinese (zh)
Other versions
CN108399362A (en
Inventor
林倞
尹森堂
张冬雨
王青
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sun Yat Sen University
Original Assignee
Sun Yat Sen University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sun Yat Sen University filed Critical Sun Yat Sen University
Priority to CN201810069322.XA priority Critical patent/CN108399362B/en
Priority to PCT/CN2018/095058 priority patent/WO2019144575A1/en
Publication of CN108399362A publication Critical patent/CN108399362A/en
Application granted granted Critical
Publication of CN108399362B publication Critical patent/CN108399362B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0495—Quantised networks; Sparse networks; Compressed networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Human Computer Interaction (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

本发明公开了一种快速行人检测方法及装置，所述方法包括如下步骤：步骤S1，构建可配置的基于卷积神经网络的深度模型，利用训练样本学习出构建的网络参数，获得用于测试过程的模型；步骤S2，输入测试样本，通过训练好的模型利用神经网络感知域的变化规律使用不同的中间层对不同尺度范围内的目标物体进行检测，预测出图像中目标物体的框图，本发明通过利用神经网络感知域的变化规律，使用不同的中间层对特定尺度范围内的目标物体进行检测，更好的适应了感知域与物体大小的关系，有效提高了检测结果。

Figure 201810069322

The invention discloses a fast pedestrian detection method and device. The method includes the following steps: Step S1, constructing a configurable deep model based on a convolutional neural network, using training samples to learn the constructed network parameters, and obtaining parameters for testing. The model of the process; step S2, input the test sample, use the trained model to use the change rule of the neural network perception domain to use different intermediate layers to detect the target objects in different scales, and predict the block diagram of the target object in the image. The invention uses different intermediate layers to detect target objects within a specific scale range by using the changing law of the neural network perception domain, which better adapts to the relationship between the perception domain and the size of the object, and effectively improves the detection result.

Figure 201810069322

Description

Rapid pedestrian detection method and device
Technical Field
The invention relates to the technical field of pedestrian detection, in particular to a rapid pedestrian detection method and device facing an embedded system based on deep learning.
Background
As part of target detection in computer vision, pedestrian detection has important significance in real world application, more and more cameras are deployed in public places with the maturity of image acquisition technology and the reduction of storage technology cost, and on the other hand, with the implementation of automatic driving and intelligent transportation, the vehicle-mounted camera also generates massive video resources. Traditional manual screening and processing not only has low efficiency and consumes a large amount of manpower and material resources, but also may introduce some human factors to cause some deviations. In recent years, deep learning has made an unprecedented breakthrough in the field of computer vision, and not only is efficiency far better than manpower, but also accuracy is better than that of human beings in many fields. Therefore, the problem of effectively using the deep learning method to detect pedestrians is receiving attention.
People are one of the most important targets in video surveillance or automatic driving, and the primary task of pedestrian detection is to identify the presence of human bodies and provide corresponding labeled information. The quality of images captured in the real world is uneven, detection of small objects and sheltered objects is always a difficulty in pedestrian detection, on the other hand, some fuzzy images are often captured by the vehicle-mounted camera, and a large number of objects similar to pedestrians but not pedestrians exist in the images. Particularly, in the embedded system, because the large neural network model with strong recognition capability is usually difficult to efficiently run on the embedded device with limited computing resources, and the application requirements of the embedded device are real-time, it is important to consider the detection accuracy and efficiency to be fast pedestrian detection for the embedded system.
Disclosure of Invention
In order to overcome the defects in the prior art, an object of the present invention is to provide a method and an apparatus for rapid pedestrian detection, which utilize the change rule of the neural network sensing domain and use different intermediate layers to detect a target object within a specific scale range, so as to better adapt to the relationship between the sensing domain and the object size and effectively improve the detection result.
The invention also aims to provide a rapid pedestrian detection method and a rapid pedestrian detection device, which can obtain an squeeze VGG-16 network which meets the requirements of an embedded system by adjusting and training the VGG-16 network, effectively reduce the parameter quantity of a network model and accelerate the calculation efficiency.
Another objective of the present invention is to provide a method and an apparatus for rapid pedestrian detection, which amplify a feature map of a specific network layer by a deconvolution method, so as to enhance the detection of small objects, and hardly increase the video memory and the calculation amount compared with the conventional image amplification method.
It is still another object of the present invention to provide a method and apparatus for rapid pedestrian detection, which has excellent performance for the detection of fuzzy objects and small distant objects by using a region 1.5 times as large as the target object as background semantic features added to the network.
In order to achieve the above and other objects, the present invention provides a rapid pedestrian detection method, comprising the steps of:
step S1, constructing a configurable depth model based on a convolutional neural network, and learning constructed network parameters by using training samples to obtain a model for a test process;
and step S2, inputting a test sample, detecting the target object in different scale ranges by using different intermediate layers through the trained model by using the change rule of the neural network perception domain, and predicting a block diagram of the target object in the image.
Preferably, the step S1 further includes:
constructing a configurable depth model based on a convolutional neural network;
inputting a training sample;
initializing a convolutional neural network and parameters thereof, including weights and offsets of each layer connection in the network layer;
and learning the constructed network parameters, namely the model for the test process by using the training samples by adopting a forward propagation algorithm and a backward propagation algorithm.
Preferably, the depth model includes a multi-scale target candidate network and a target detection network, and the target candidate network respectively generates candidate block diagrams for target objects of different scales in an intermediate layer based on differences of features proposed by different layers of a convolutional neural network; and the target detection network carries out refined classification and detection on the basis of the candidate block diagram output by the target candidate network.
Preferably, the convolutional neural network is formed by stacking a convolutional layer, a down-sampling layer and an up-sampling layer. The convolutional layer is used for performing convolution operation on an input image or a characteristic diagram on a two-dimensional space and extracting hierarchical characteristics; the down-sampling layer uses a non-overlapped max-firing operation, and the operation is used for extracting the features with unchanged shapes and offsets, reducing the size of a feature map and improving the calculation efficiency; the upsampling layer is an operation of deconvolving the input feature map on a two-dimensional space, so as to increase pixels of the feature map.
Preferably, the depth model adopts a Squeeze VGG-16 convolutional neural network as a backbone network, and the Squeeze VGG-16 convolutional neural network adopts a network structure with a conv1-1 layer and a 12-layer Fire module layer which is immediately followed as a feature extraction layer.
Preferably, the target candidate network generates network branches at Fire9, Fire12, conv6 and added posing layers according to convolutional layer characteristics on the basis of the Squeeze VGG-16 convolutional neural network so as to perform regression of candidate frames of the detected object with different scales.
Preferably, on the basis of the target candidate region, the target detection network takes a picture region with a preset multiple size of the target candidate region as background semantic information of the target, performs primary up-sampling on a feature map of a Fire9 layer as information for enhancing small object perception, performs pooling of the background semantic information and the up-sampling information in the region of interest to obtain features with a fixed size, and then adds a full connection layer to perform regression of the category and the final candidate frame.
Preferably, the training sample includes RGB image data and labeling information of a pedestrian region in the image, and the image data for actual training is a small patch cut according to the region where the pedestrian is located.
Preferably, the back propagation algorithm needs to first find the loss function between the target block diagram of the forward propagation prediction and the actual target block diagram of the image
Figure BDA0001557667470000031
Then, the gradient of the parameter W is obtained, and the gradient descending algorithm is adopted to update W so as to minimize the loss function
Figure BDA0001557667470000032
Assuming that the middle layer has M branches to output the target candidate region, lmRepresenting the loss function of branch m, αmIs represented bymWeight of function, S ═ S1，S2，…，SMDenotes the target object of corresponding scale, then the loss function
Figure BDA0001557667470000033
Can be defined as:
Figure BDA0001557667470000041
to achieve the above object, the present invention further provides a rapid pedestrian detection system, including:
the training unit is used for constructing a configurable depth model based on the convolutional neural network, and learning constructed network parameters by using training samples to obtain a model for a testing process;
and the detection unit is used for inputting a test sample, detecting the target object in different scale ranges by using different intermediate layers through the trained model by utilizing the change rule of the neural network perception domain, and predicting a block diagram of the target object in the image.
Compared with the prior art, the rapid pedestrian detection method and the rapid pedestrian detection device provided by the invention use a method of compressing a network for reference, adjust and train the network of the VGG-16 to obtain the squeeze VGG-16 network which meets the requirements of an embedded system, effectively reduce the parameter quantity of a network model and accelerate the calculation efficiency; on the other hand, aiming at the problem that the sensing domain is not consistent with the size of the object in the traditional detection method, the invention utilizes the change rule of the neural network sensing domain (namely, the deeper the neural network layer is, the larger the sensing domain is, the larger the object is suitable for detecting the larger object), uses different intermediate layers to detect the object in the specific scale range, better adapts to the relation between the sensing domain and the object size, and effectively improves the detection result; in addition, in order to enhance the detection of small objects, the characteristic diagram of a specific network layer is amplified by using a deconvolution method, and compared with the traditional image amplification method, the method hardly increases the video memory and the calculation amount; in order to enhance the detection of the fuzzy object, a region with the size 1.5 times that of the target object is used as a background semantic feature to be added into the network on the feature map of the layer, and the detection of the fuzzy object and the long-distance small object has excellent performance.
Drawings
FIG. 1 is a flow chart of the steps of a rapid pedestrian detection method of the present invention;
FIG. 2 is a schematic diagram of the structure of the Squeeze VGG-16 neural network according to the embodiment of the present invention;
FIG. 3 is a diagram of a Fire module in accordance with an embodiment of the present invention;
FIG. 4 is a schematic diagram of a target candidate network according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a target detection network according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of a rapid pedestrian detection process according to an embodiment of the present invention;
FIG. 7 is a system architecture diagram of a rapid pedestrian detection device in accordance with the present invention;
FIG. 8 is a detailed block diagram of a training unit in accordance with an embodiment of the present invention;
FIG. 9 is a detailed structure diagram of a detecting unit according to an embodiment of the present invention.
Detailed Description
Other advantages and capabilities of the present invention will be readily apparent to those skilled in the art from the present disclosure by describing the embodiments of the present invention with specific embodiments thereof in conjunction with the accompanying drawings. The invention is capable of other and different embodiments and its several details are capable of modification in various other respects, all without departing from the spirit and scope of the present invention.
Fig. 1 is a flow chart illustrating steps of a rapid pedestrian detection method according to the present invention. As shown in fig. 1, the present invention provides a rapid pedestrian detection method, which comprises the following steps:
and step S1, constructing a configurable depth model based on the convolutional neural network, and learning constructed network parameters by using training samples to obtain a model for the test process. In a specific embodiment of the present invention, the depth model is composed of two sub-networks: the first sub-network is a multi-scale target candidate network and is used for extracting character features and giving out candidate regions, and specifically, the target candidate network respectively generates candidate block diagrams of pedestrians in different scales in an intermediate layer based on differences of the characteristics given out by different layers of a convolutional neural network; and the second sub-network is a target detection network, enhances the detection effect, shares parameters with the target candidate network, and performs refined classification and detection on the basis of the candidate block diagram. Specifically, step S1 further includes:
and S100, constructing a configurable depth model based on the convolutional neural network.
The convolutional neural network is formed by stacking a convolutional layer, a down-sampling layer and an up-sampling layer, wherein the convolutional layer is used for performing convolution operation on an input image or a characteristic diagram in a two-dimensional space and extracting hierarchical characteristics; the down-sampling layer uses a non-overlapped max-firing operation, and the operation is used for extracting the features with unchanged shapes and offsets, reducing the size of a feature map and improving the calculation efficiency; the upsampling layer is an operation of deconvolving an input feature map in a two-dimensional space, is used for increasing pixels of the feature map, and is mainly used for a target detection network and improving a detection effect, in the specific embodiment of the invention, an Squeeze VGG-16 convolutional neural network is used as a backbone network, as shown in FIG. 2, the Squeeze VGG-16 convolutional neural network adopts a conv1-1 layer and a 12-layer Fire module which is immediately followed as a convolutional layer for extracting features; wherein pool1-pool5 is a down-sampling layer; a pre-trained model on the ImageNet dataset was used as initialization. Namely, the invention firstly trains Squeeze VGG-16 in advance by using ImageNet data set as network initialization.
Fig. 3 is a schematic structural diagram of a Fire module in an embodiment of the present invention. As shown in fig. 3, the Fire module is composed of two convolutional layers with convolution kernel size of 1 × 1 and one convolutional layer with convolution kernel size of 3 × 3, and aims to replace the convolution kernel of 3 × 3 with convolution kernel of 1 × 1, so as to reduce the parameter amount by 9 times, but in order not to affect the characterization capability of the network, the convolution kernel of 1 × 1 is not replaced completely, and the convolution kernel of 3 × 3 is used partially, so that another benefit of this is to reduce the input channel of convolution kernel of 3 × 3, and at the same time, to achieve the effect of reducing the parameter amount, specifically, the Fire module firstly uses convolution kernel of 1 × 1 to perform dimensionality reduction operation on the input layer, then uses convolution kernel of 1 × 1 and convolution kernel of 3 to extract the features, and finally connects the two parts of features, in such a way, the computation amount and the model parameters are greatly reduced.
Fig. 4 is a schematic diagram of a target candidate network according to an embodiment of the present invention. In the embodiment of the invention, the target candidate network generates network branches on the basis of the Squeeze VGG-16 convolutional neural network, wherein the network branches are calculated by 4 layers in total on the basis of the convolutional layer characteristics, namely Fire9, Fire12, conv6 and added firing layers, and the network branches perform regression on candidate frames of objects detected in different scales. However, for the Fire-9 layer, which is closer to the lower layer of the backbone network, the influence on the gradient is larger than that of other layers, the learning process is unstable, and therefore, a buffer layer is added, as shown in the det-conv layer in fig. 4, and the buffer layer prevents the gradient of the detection branch from being directly back-propagated to the backbone layer.
The invention utilizes the change rule of the neural network perception domain (namely, the deeper the neural network layer, the larger the perception domain, and is suitable for detecting larger target objects), and uses different intermediate layers to detect the target objects in a specific scale range, thereby better adapting to the relationship between the perception domain and the object size and effectively improving the detection result.
Fig. 5 is a schematic diagram of an architecture of a target detection network according to an embodiment of the present invention. The target detection network and the target candidate network share parameters, and candidate frames of the target candidate network are summarized to enhance the distinguishing capability of the monitoring network on objects and backgrounds. In the specific embodiment of the present invention, the target detection network uses, on the basis of the target candidate region, a picture region 1.5 times as large as the target candidate region as background semantic information of the target; performing primary up-sampling on a feature map of a Fire9 layer to serve as information for enhancing small object perception, performing pooling (ROI posing) of a region of interest on background semantic information and up-sampling information to obtain features with fixed sizes, then adding a layer of full connection layer, and performing regression of categories and final candidate frames, wherein specifically, a trunk cnn layer is connected with a propalss node and is used for summarizing candidate frame information obtained by a target candidate network; on the other hand, for the feature map of Fire9 layer, W and H are the width and height of the input picture, cube 1 represents the mapping of the object region on the feature map, and cube 2 represents the mapping of the context region on the feature map, the context region is about 1.5 times of the object region, and in order to enhance the detection of small objects, the Fire9 layer is up-sampled once again, and then the pooling of the region of interest is used to obtain features of fixed size, similar to the fast RCNN algorithm; the processed features of the Fire9 layer are connected (concat) with the features summarized by the prosassals, and then a fully connected layer is added to perform regression of the category and the final candidate frame, which is not described herein again.
Step S101, inputting training samples.
The training process needs to provide a corresponding frame of a reference character in the image, and meanwhile, in order to accelerate the training, the training process cuts the image containing the reference character from the original image to form patch (image block), and the patch is smaller than the original image and used for training, so that the training process is effectively accelerated. Specifically, in the present invention, the input training sample includes RGB image data and labeling information of a pedestrian region in an image, and the image data for actual training is a small patch (image block) cut out according to the region where a pedestrian is present. Expressed in mathematical language, training samples
Figure BDA0001557667470000071
Wherein XiA patch representing a training picture; in practical applications, there are other categories than the category of pedestrian, such as K categories of background, cyclist, sitting person, etc., so the label data Yi＝(yi，bi) By category label yiE {0, 1, 2.,. K } and block diagram coordinate points
Figure BDA0001557667470000072
Is composed of (a) wherein
Figure BDA0001557667470000073
Is the starting coordinate point in the upper left corner of the diagram,
Figure BDA0001557667470000074
the frame width and height.
Step S102, initializing the convolutional neural network and parameters thereof, including the weight and the bias of each layer connection in the network layer. Specifically, the present invention utilizes ImageNet data set to pre-train Squeeze VGG-16 convolutional neural networks as network initialization.
Step S103, learning the constructed network parameters, namely the model for the test process by using the training samples by adopting a forward propagation algorithm and a backward propagation algorithm.
In the invention, the forward propagation algorithm firstly normalizes the size of an input image to be 3 × 480 × 640, cuts out patch with the size of 3 × 448 × 448 and corresponding labeled information as the input of a convolutional neural network, passes through a convolutional Layer, a down-sampling Layer and a correction linear unit Layer (ReLU nonlinear Layer), and has the image characteristic diagram size of 512 × 60 × 80 at a Fire9 Layer; at the Fire12 level, the feature map size is 512 × 30 × 40, and the feature map sizes of the two latter branches are 512 × 15 × 20 and 512 × 8 × 10 in this order. On different feature maps, four coordinate points and category information of a target block diagram are obtained in a convolution mode, for example, a Fire9 layer is taken as an example, and assuming that only pedestrians and backgrounds are detected, the output is that the feature size is 6 × 60 × 80, wherein 6 includes four coordinate points of the background, two categories of pedestrians and a candidate block diagram. In the target detection network, the candidate block diagrams obtained by each branch layer are collected at a prosals node, and are simultaneously superposed with the background semantic information of the Fire9 layer and the characteristics obtained by pooling the up-sampling information in the region of interest to perform final block diagram regression and category regression.
In the present invention, the backward propagation algorithm needs to first find the loss function between the target block diagram of forward (i.e. forward) propagation prediction and the actual target block diagram of image
Figure BDA0001557667470000081
Then, the gradient of the parameter W is obtained, and the gradient descending algorithm is adopted to update W so as to minimize the loss function
Figure BDA0001557667470000082
Assuming that there are M branches in the middle layer to output target candidate regions (all target objects in the image can be approximately detected by the perception domains of M scales), lmRepresenting the loss function of branch m, αmIs represented bymWeight of function, S ═ S1，S2，…，SMDenotes the target object of corresponding scale, then the loss function
Figure BDA0001557667470000083
Can be defined as:
Figure BDA0001557667470000084
the loss function, for a specific detection layer m, only contributes to the loss function if the target scale is within the range that m can detect, so the loss function is defined as
Figure BDA0001557667470000085
Wherein p (X) ═ p0(X)，...，pK(X)) represents a probability distribution of the target class; λ is the equilibrium coefficient; b are 4 coordinate points of the block diagram,
Figure BDA0001557667470000086
pointing to coordinate points obtained by forward propagation; in the loss function, a cross-entropy loss function is used to define class regression, i.e.
Lcls(p(X)，y)＝-logy(P(X)) (3)
Regression of the target block diagram was performed using the smoothed Manhattan distance criterion (smooth L1 criterion), defined as follows
Figure BDA0001557667470000091
And step S2, detecting the target objects in different scale ranges by using different intermediate layers through the trained model and utilizing the change rule of the neural network perception domain, and predicting a block diagram of the target objects (such as pedestrians) in the image.
Specifically, step S2 further includes:
step S200, loading the trained model;
step S201, inputting a test sample;
and S202, detecting pedestrians in different scale ranges by using different intermediate layers through the change rule of the neural network perception domain by using the trained model, and predicting a block diagram of the pedestrians in the image. Fig. 6 is a schematic diagram of a process of rapid pedestrian detection in an embodiment of the present invention, that is, a target candidate network in a model is used to generate network branches in 4 total layers of fire9, fire12, conv6 and added pooling layers on the basis of the Squeeze VGG-16 convolutional neural network according to the characteristics of convolutional layers, and target candidate regions (middle layer a, middle layer b, middle layer c) of an object are detected in different scales; then, by using a target detection network, on the basis of a target candidate region, taking a picture region 1.5 times as large as the target candidate region as background semantic information of a target, performing primary up-sampling on a feature map of a Fire9 layer as information for enhancing small object perception, pooling the background semantic information and the up-sampling information in a region of interest to obtain features of a fixed size, then adding a full connection layer, and performing regression of categories and final candidate frames. Preferably, in step S202, the feature map of the specific network layer is further enlarged by using a deconvolution method.
The pedestrian detection method provided by the invention respectively uses evaluation indexes in two aspects: average precision rate mAP and frames per second FPS. The mAP is used for evaluating the condition of the intersection ratio of the final detection area and the real target person area, and the average value of the precision ratio under different intersection ratios; FPS, which is mainly an efficiency indicator, refers to the number of pictures that can be processed per second.
Fig. 7 is a system architecture diagram of a rapid pedestrian detection device according to the present invention. As shown in fig. 7, the present invention provides a rapid pedestrian detection apparatus, including:
and the training unit 70 is used for constructing a configurable depth model based on the convolutional neural network, and learning constructed network parameters by using training samples to obtain a model for the test process. In an embodiment of the present invention, the depth model constructed by the training unit 70 is composed of two sub-networks: the first sub-network is a multi-scale target candidate network and is used for extracting character features and giving out candidate regions, and specifically, the target candidate network respectively generates candidate block diagrams of pedestrians in different scales in an intermediate layer based on differences of the characteristics given out by different layers of a convolutional neural network; and the second sub-network is a target detection network, enhances the detection effect, shares parameters with the target candidate network, and performs refined classification and detection on the basis of the candidate block diagram. Specifically, as shown in fig. 8, the training unit 70 further includes:
and the model building unit 701 is used for building a configurable depth model based on the convolutional neural network.
The convolutional neural network is formed by stacking a convolutional layer, a down-sampling layer and an up-sampling layer, wherein the convolutional layer is used for performing convolution operation on an input image or a characteristic diagram in a two-dimensional space and extracting hierarchical characteristics; the down-sampling layer uses non-overlapping max-forcing operation, which is used for extracting features with unchanged shapes and offsets, reducing the size of a feature map and improving the calculation efficiency, and the up-sampling layer is used for performing deconvolution operation on an input feature map on a two-dimensional space to increase pixels of the feature map. In a specific embodiment of the invention, an Squeeze VGG-16 convolutional neural network is adopted as a backbone network.
In the embodiment of the invention, the target candidate network generates network branches on the basis of the Squeeze VGG-16 convolutional neural network, wherein the network branches are calculated by 4 layers in total on fire9, fire12, conv6 and added posing layers according to the characteristics of convolutional layers, and the network branches perform regression on candidate frames of objects detected by different scales. However, for the fire-9 layer, the gradient is relatively close to the lower layer of the backbone network, the influence on the gradient is large compared with other layers, the learning process is unstable, and therefore, a buffer layer is added, and the buffer layer prevents the gradient of the detection branch from being directly back-propagated to the backbone layer.
The target detection network and the target candidate network share parameters, and candidate frames of the target candidate network are summarized to enhance the distinguishing capability of the monitoring network on objects and backgrounds. In the specific embodiment of the present invention, the target detection network uses, on the basis of the target candidate region, a picture region 1.5 times as large as the target candidate region as background semantic information of the target; the method comprises the steps of performing primary up-sampling on a feature map of a Fire9 layer to serve as information for enhancing perception of small objects, performing pooling of background semantic information and up-sampling information in a region of interest to obtain features of a fixed size, then adding a layer of full connection layer to perform regression of categories and final candidate frames, specifically, connecting a backbone cnn layer with a subnetwork of a prosal, wherein W and H are the width and the height of an input picture, a cube 1 represents the firing of an object region, a cube 2 represents the firing of a context region, the context region is about 1.5 times of the object region, performing primary up-sampling on a Fire9 layer to enhance detection of the small objects, then obtaining the features of the fixed size by using pooling of the region of interest similar to a false RCNN algorithm, and then adding a layer of full connection layer to perform regression of the categories and the final candidate frames.
A training sample input unit 702 for inputting training samples.
Specifically, training samples
Figure BDA0001557667470000111
Wherein XiOne patch representing a training picture, labeled data Yi＝(yi，bi) By category label yiAnd block diagram coordinate points
Figure BDA0001557667470000112
And (4) forming.
The initialization unit 703 is configured to initialize the convolutional neural network and its parameters, including the weight and the offset of each layer connection in the network layer. Specifically, the present invention utilizes ImageNet data set to pre-train Squeeze VGG-16 convolutional neural networks as network initialization.
And the sample training unit 704 is used for learning the constructed network parameters, namely the model for the test process by using the training samples by adopting a forward propagation algorithm and a backward propagation algorithm.
In the invention, the forward propagation algorithm firstly normalizes the size of an input image to be 3 × 480 × 640, cuts out patch with the size of 3 × 448 × 448 and corresponding labeled information as the input of a convolutional neural network, passes through a convolutional Layer, a down-sampling Layer and a correction linear unit Layer (ReLU nonlinear Layer), and has the image characteristic diagram size of 512 × 60 × 80 at a Fire9 Layer; at the Fire12 level, the feature map size is 512 × 30 × 40, and the feature map sizes of the two latter branches are 512 × 15 × 20 and 512 × 8 × 10 in this order. On different feature maps, four coordinate points and category information of a target block diagram are obtained in a convolution mode, for example, a Fire9 layer is taken as an example, and assuming that only pedestrians and backgrounds are detected, the output is that the feature size is 6 × 60 × 80, wherein 6 includes four coordinate points of the background, two categories of pedestrians and a candidate block diagram. In the target detection network, the candidate block diagrams obtained by each branch layer are collected at a prosals node, and are simultaneously superposed with the background semantic information of the Fire9 layer and the characteristics obtained by pooling the up-sampling information in the region of interest to perform final block diagram regression and category regression.
The backward propagation algorithm needs to firstly calculate the loss function of the target block diagram of the forward propagation prediction and the actual target block diagram of the image
Figure BDA0001557667470000121
Then, the gradient of the parameter W is obtained, and the gradient descending algorithm is adopted to update W so as to minimize the loss function
Figure BDA0001557667470000126
Assuming that there are M branches in the middle layer to output target candidate regions (all target objects in the image can be approximately detected by the perception domains of M scales), lmRepresenting the loss function of branch m, αmIs represented bymWeight of function, S ═ S1，S2，…，SMDenotes the target object of corresponding scale, then the loss function
Figure BDA0001557667470000122
Can be defined as:
Figure BDA0001557667470000123
the loss function, for a specific detection layer m, only contributes to the loss function if the target scale is within the range that m can detect, so the loss function is defined as
Figure BDA0001557667470000124
Wherein p (X) ═ p0(X)，...，pK(X)) is the probability distribution of the target class. In the loss function, a cross-entropy loss function is used to define class regression, i.e.
Lcls(p(X)，y)＝-logy(P(X))
Regression of the target block diagram was performed using smooth L1 criterion, defined as follows
Figure BDA0001557667470000125
The detection unit 71 is configured to input a test sample, detect a target object (e.g., a pedestrian) in a range of different scales by using different intermediate layers according to a change rule of a neural network perception domain through a trained model, and predict a block diagram of the target object (e.g., the pedestrian) in the image.
Specifically, as shown in fig. 9, the detection unit 71 further includes:
a model loading unit 710 for loading the trained model;
a test sample input unit 711 for inputting a test sample;
and the image prediction unit 712 is configured to detect pedestrians in different scale ranges by using the trained model and using different intermediate layers according to the change rule of the neural network sensing domain through the trained model, and predict a block diagram of the pedestrian in the image. Specifically, the image prediction unit 712 performs different-scale detection on the target candidate area of the object by using the target candidate network in the model and generating network branches in 4 layers in total of Fire9, Fire12, conv6 and added posing layers according to the characteristics of the convolution layer on the basis of the Squeeze VGG-16 convolutional neural network; then, by using a target detection network, on the basis of a target candidate region, taking a picture region 1.5 times as large as the target candidate region as background semantic information of a target, performing primary up-sampling on a feature map of a Fire9 layer as information for enhancing small object perception, pooling the background semantic information and the up-sampling information in a region of interest to obtain features of a fixed size, then adding a full connection layer, and performing regression of categories and final candidate frames.
In conclusion, the rapid pedestrian detection method and the rapid pedestrian detection device provided by the invention use a method of compressing a network for reference, adjust and train the network of the VGG-16 to obtain the squeeze VGG-16 network which meets the requirements of an embedded system, effectively reduce the parameters of a network model and accelerate the calculation efficiency; on the other hand, aiming at the problem that the sensing domain is not consistent with the size of the object in the traditional detection method, the invention utilizes the change rule of the neural network sensing domain (namely, the deeper the neural network layer is, the larger the sensing domain is, the larger the object is suitable for detecting the larger object), uses different intermediate layers to detect the object in the specific scale range, better adapts to the relation between the sensing domain and the object size, and effectively improves the detection result; in addition, in order to enhance the detection of small objects, the characteristic diagram of a specific network layer is amplified by using a deconvolution method, and compared with the traditional image amplification method, the method hardly increases the video memory and the calculation amount; in order to enhance the detection of the fuzzy object, a region with the size 1.5 times that of the target object is used as a background semantic feature to be added into the network on the feature map of the layer, and the detection of the fuzzy object and the long-distance small object has excellent performance.
The foregoing embodiments are merely illustrative of the principles and utilities of the present invention and are not intended to limit the invention. Modifications and variations can be made to the above-described embodiments by those skilled in the art without departing from the spirit and scope of the present invention. Therefore, the scope of the invention should be determined from the following claims.

Claims (4)

1. A rapid pedestrian detection method comprises the following steps:
step S1, constructing a configurable depth model based on a convolutional neural network, and learning constructed network parameters by using training samples to obtain a model for a test process;
step S2, inputting a test sample, detecting target objects in different scale ranges by using different intermediate layers through a trained model and utilizing the change rule of a neural network perception domain, and predicting a block diagram of the target objects in the image;
step S1 further includes:
constructing a configurable depth model based on a convolutional neural network;
inputting a training sample;
initializing a convolutional neural network and parameters thereof, including weights and offsets of each layer connection in the network layer;
learning the constructed network parameters, namely a model for the test process by using a training sample by adopting a forward propagation algorithm and a backward propagation algorithm;
the depth model comprises a multi-scale target candidate network and a target detection network, wherein the target candidate network respectively generates candidate block diagrams of target objects with different scales in the middle layer based on the difference of features proposed by different layers of the convolutional neural network; the target detection network carries out refined classification and detection on the basis of the candidate block diagram output by the target candidate network;
the convolutional neural network is formed by stacking a convolutional layer, a down-sampling layer and an up-sampling layer, wherein the convolutional layer is used for performing convolution operation on an input image or a characteristic diagram in a two-dimensional space and extracting hierarchical characteristics; the down-sampling layer uses a non-overlapped max-firing operation, and the operation is used for extracting the features with unchanged shapes and offsets, reducing the size of a feature map and improving the calculation efficiency; the upsampling layer is an operation of deconvolving the input feature map on a two-dimensional space, so as to increase pixels of the feature map.
2. The rapid pedestrian detection method of claim 1, wherein: the depth model adopts a Squeeze VGG-16 convolutional neural network as a backbone network, and the Squeeze VGG-16 convolutional neural network adopts a network structure with a conv1-1 layer and a 12-layer Fire module layer which is immediately followed as features for extraction.
3. A rapid pedestrian detection method according to claim 2, characterized in that: and the target candidate network generates network branches on the basis of the Squeeze VGG-16 convolutional neural network and in the Fire9, the Fire12, the conv6 and the added posing layer according to the convolutional layer characteristics so as to carry out regression on candidate frames of the detected object with different scales.
4. A rapid pedestrian detection method according to claim 2, characterized in that: the target detection network takes a picture area with the size of a preset multiple of the target candidate area as background semantic information of a target on the basis of the target candidate area, performs primary up-sampling on a feature map of a Fire9 layer to serve as information for enhancing small object perception, performs pooling of the background semantic information and the up-sampling information in an interested area to obtain features with fixed size, and then adds a full connection layer to perform regression of categories and final candidate frames.
CN201810069322.XA 2018-01-24 2018-01-24 Rapid pedestrian detection method and device Active CN108399362B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201810069322.XA CN108399362B (en) 2018-01-24 2018-01-24 Rapid pedestrian detection method and device
PCT/CN2018/095058 WO2019144575A1 (en) 2018-01-24 2018-07-10 Fast pedestrian detection method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810069322.XA CN108399362B (en) 2018-01-24 2018-01-24 Rapid pedestrian detection method and device

Publications (2)

Publication Number Publication Date
CN108399362A CN108399362A (en) 2018-08-14
CN108399362B true CN108399362B (en) 2022-01-07

Family

ID=63094281

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810069322.XA Active CN108399362B (en) 2018-01-24 2018-01-24 Rapid pedestrian detection method and device

Country Status (2)

Country Link
CN (1) CN108399362B (en)
WO (1) WO2019144575A1 (en)

Families Citing this family (360)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018176000A1 (en) 2017-03-23 2018-09-27 DeepScale, Inc. Data synthesis for autonomous control systems
US11893393B2 (en) 2017-07-24 2024-02-06 Tesla, Inc. Computational array microprocessor system with hardware arbiter managing memory requests
US11409692B2 (en) 2017-07-24 2022-08-09 Tesla, Inc. Vector computational unit
US11157441B2 (en) 2017-07-24 2021-10-26 Tesla, Inc. Computational array microprocessor system using non-consecutive data formatting
US10671349B2 (en) 2017-07-24 2020-06-02 Tesla, Inc. Accelerated mathematical engine
US12307350B2 (en) 2018-01-04 2025-05-20 Tesla, Inc. Systems and methods for hardware-based pooling
US11561791B2 (en) 2018-02-01 2023-01-24 Tesla, Inc. Vector computational unit receiving data elements in parallel from a last row of a computational array
US11215999B2 (en) 2018-06-20 2022-01-04 Tesla, Inc. Data pipeline and deep learning system for autonomous driving
US11361457B2 (en) 2018-07-20 2022-06-14 Tesla, Inc. Annotation cross-labeling for autonomous control systems
US11636333B2 (en) 2018-07-26 2023-04-25 Tesla, Inc. Optimizing neural network structures for embedded systems
CN109089040B (en) * 2018-08-20 2021-05-14 Oppo广东移动通信有限公司 Image processing method, image processing device and terminal device
US11562231B2 (en) 2018-09-03 2023-01-24 Tesla, Inc. Neural networks for embedded devices
KR102813985B1 (en) 2018-10-11 2025-05-29 테슬라, 인크. Systems and methods for training machine models with augmented data
CN109409364A (en) * 2018-10-16 2019-03-01 北京百度网讯科技有限公司 Image annotation method and device
US11196678B2 (en) 2018-10-25 2021-12-07 Tesla, Inc. QOS manager for system on a chip communications
CN109508675B (en) * 2018-11-14 2020-07-28 广州广电银通金融电子科技有限公司 Pedestrian detection method for complex scene
CN109522855B (en) * 2018-11-23 2020-07-14 广州广电银通金融电子科技有限公司 Low-resolution pedestrian detection method, system and storage medium combining ResNet and SENet
CN109522966B (en) * 2018-11-28 2022-09-27 中山大学 A target detection method based on densely connected convolutional neural network
US11816585B2 (en) 2018-12-03 2023-11-14 Tesla, Inc. Machine learning models operating at different frequencies for autonomous vehicles
US11537811B2 (en) 2018-12-04 2022-12-27 Tesla, Inc. Enhanced object detection for autonomous vehicles based on field view
CN109670439A (en) * 2018-12-14 2019-04-23 中国石油大学(华东) A kind of pedestrian and its location detection method end to end
CN109685718B (en) * 2018-12-17 2020-11-10 中国科学院自动化研究所 Picture squaring zooming method, system and device
CN109886066B (en) * 2018-12-17 2023-05-09 南京理工大学 Rapid target detection method based on multi-scale and multi-layer feature fusion
US11610117B2 (en) 2018-12-27 2023-03-21 Tesla, Inc. System and method for adapting a neural network model on a hardware platform
CN109902800B (en) * 2019-01-22 2020-11-27 北京大学 A method for detecting general objects based on a multi-level backbone network based on quasi-feedback neural network
US11150664B2 (en) 2019-02-01 2021-10-19 Tesla, Inc. Predicting three-dimensional features for autonomous driving
US10997461B2 (en) 2019-02-01 2021-05-04 Tesla, Inc. Generating ground truth for machine learning from time series elements
CN111523351A (en) * 2019-02-02 2020-08-11 北京地平线机器人技术研发有限公司 Neural network training method and device and electronic equipment
US11567514B2 (en) 2019-02-11 2023-01-31 Tesla, Inc. Autonomous and user controlled vehicle summon to a target
US10956755B2 (en) 2019-02-19 2021-03-23 Tesla, Inc. Estimating object properties using visual image data
WO2020168515A1 (en) * 2019-02-21 2020-08-27 深圳市大疆创新科技有限公司 Image processing method and apparatus, image capture processing system, and carrier
CN109993712B (en) 2019-04-01 2023-04-25 腾讯科技（深圳）有限公司 Image processing model training method, image processing method and related equipment
CN110110783A (en) * 2019-04-30 2019-08-09 天津大学 A kind of deep learning object detection method based on the connection of multilayer feature figure
CN110110793B (en) * 2019-05-10 2021-10-26 中山大学 Binocular image rapid target detection method based on double-current convolutional neural network
CN118196828B (en) * 2019-06-06 2025-10-21 华为技术有限公司 Object recognition method and device
CN110659664B (en) * 2019-08-02 2022-12-13 杭州电子科技大学 A method for recognizing small objects with high precision based on SSD
CN110633631B (en) * 2019-08-06 2022-02-18 厦门大学 Pedestrian re-identification method based on component power set and multi-scale features
CN110619268B (en) * 2019-08-07 2022-11-25 北京市新技术应用研究所 Pedestrian re-identification method and device based on space-time analysis and depth features
CN110533084B (en) * 2019-08-12 2022-09-30 长安大学 Multi-scale target detection method based on self-attention mechanism
CN110473195B (en) * 2019-08-13 2023-04-18 中山大学 Medical focus detection framework and method capable of being customized automatically
CN110427915B (en) * 2019-08-14 2022-09-27 北京百度网讯科技有限公司 Method and apparatus for outputting information
CN110705583B (en) * 2019-08-15 2024-03-15 平安科技（深圳）有限公司 Cell detection model training method, device, computer equipment and storage medium
CN110490252B (en) * 2019-08-19 2022-11-15 西安工业大学 Indoor people number detection method and system based on deep learning
CN110580726B (en) * 2019-08-21 2022-10-04 中山大学 Dynamic convolution network-based face sketch generation model and method in natural scene
CN110659576A (en) * 2019-08-23 2020-01-07 深圳久凌软件技术有限公司 Pedestrian searching method and device based on joint judgment and generation learning
CN110647816B (en) * 2019-08-26 2022-11-22 合肥工业大学 Target detection method for real-time monitoring of goods shelf medicines
CN110580727B (en) * 2019-08-27 2023-04-18 天津大学 Depth V-shaped dense network imaging method with increased information flow and gradient flow
CN110675309A (en) * 2019-08-28 2020-01-10 江苏大学 An Image Style Transfer Method Based on Convolutional Neural Network and VGGNet16 Model
CN112446376B (en) * 2019-09-05 2023-08-01 中国科学院沈阳自动化研究所 Intelligent segmentation and compression method for industrial image
CN110728186B (en) * 2019-09-11 2023-04-07 中国科学院声学研究所南海研究站 Fire detection method based on multi-network fusion
CN110619676B (en) * 2019-09-18 2023-04-18 东北大学 End-to-end three-dimensional face reconstruction method based on neural network
CN110619365B (en) * 2019-09-18 2023-09-12 苏州经贸职业技术学院 Method for detecting falling water
CN110619309B (en) * 2019-09-19 2023-07-18 天地伟业技术有限公司 Embedded platform face detection method based on octave convolution and YOLOv3
CN110659601B (en) * 2019-09-19 2022-12-02 西安电子科技大学 Dense vehicle detection method for remote sensing images based on deep fully convolutional network based on central points
CN110706239B (en) * 2019-09-26 2022-11-11 哈尔滨工程大学 Scene segmentation method fusing full convolution neural network and improved ASPP module
CN110717903A (en) * 2019-09-30 2020-01-21 天津大学 Method for detecting crop diseases by using computer vision technology
CN110674777A (en) * 2019-09-30 2020-01-10 电子科技大学 An Optical Character Recognition Method in Patent Text Scenario
CN110751076B (en) * 2019-10-09 2023-03-28 上海应用技术大学 Vehicle detection method
CN110781895B (en) * 2019-10-10 2023-06-20 湖北工业大学 Image semantic segmentation method based on convolutional neural network
CN110796232A (en) * 2019-10-12 2020-02-14 腾讯科技（深圳）有限公司 Attribute prediction model training method, attribute prediction method and electronic equipment
CN110728238A (en) * 2019-10-12 2020-01-24 安徽工程大学 A Person Redetection Method Based on Fusion Neural Network
CN110728640B (en) * 2019-10-12 2023-07-18 合肥工业大学 A Double-Channel Single Image Fine Rain Removal Method
CN112668374A (en) * 2019-10-16 2021-04-16 北京灵汐科技有限公司 Image processing method and device, re-recognition network training method and electronic equipment
CN111008554B (en) * 2019-10-16 2024-02-02 合肥湛达智能科技有限公司 Deep learning-based method for identifying pedestrians without giving away in dynamic traffic zebra stripes
CN111046723B (en) * 2019-10-17 2023-06-02 安徽清新互联信息科技有限公司 Lane line detection method based on deep learning
CN110852179B (en) * 2019-10-17 2023-08-25 天津大学 Detection method of suspicious personnel intrusion based on video surveillance platform
CN111008632B (en) * 2019-10-17 2023-06-09 安徽清新互联信息科技有限公司 License plate character segmentation method based on deep learning
CN110751644B (en) * 2019-10-23 2023-05-09 上海应用技术大学 Road surface crack detection method
CN110909615B (en) * 2019-10-28 2023-03-28 西安交通大学 Target detection method based on multi-scale input mixed perception neural network
CN111008562B (en) * 2019-10-31 2023-04-18 北京城建设计发展集团股份有限公司 Human-vehicle target detection method with feature map depth fusion
CN110826476A (en) * 2019-11-02 2020-02-21 国网浙江省电力有限公司杭州供电公司 Image detection method and device for identifying target object, electronic equipment and storage medium
CN110826485B (en) * 2019-11-05 2023-04-18 中国人民解放军战略支援部队信息工程大学 Target detection method and system for remote sensing image
CN110826552A (en) * 2019-11-05 2020-02-21 华中农业大学 Grape non-destructive automatic detection device and method based on deep learning
CN110837837B (en) * 2019-11-05 2023-10-17 安徽工业大学 Vehicle violation detection method based on convolutional neural network
CN111008567B (en) * 2019-11-07 2023-03-24 郑州大学 Driver behavior identification method
CN110852272B (en) * 2019-11-11 2023-03-28 上海应用技术大学 Pedestrian detection method
CN111461160B (en) * 2019-11-11 2023-07-14 天津津航技术物理研究所 Infrared imaging seeker target tracking method for preventing cloud and fog interference
CN111222402A (en) * 2019-11-14 2020-06-02 北京理工大学 Crowd gathering density analysis method oriented to unmanned aerial vehicle image
CN111008994A (en) * 2019-11-14 2020-04-14 山东万腾电子科技有限公司 Moving target real-time detection and tracking system and method based on MPSoC
CN111126359B (en) * 2019-11-15 2023-03-28 西安电子科技大学 High-definition image small target detection method based on self-encoder and YOLO algorithm
CN111222534B (en) * 2019-11-15 2022-10-11 重庆邮电大学 Single-shot multi-frame detector optimization method based on bidirectional feature fusion and more balanced L1 loss
CN111144203B (en) * 2019-11-19 2023-06-16 浙江工商大学 Pedestrian shielding detection method based on deep learning
CN110942008B (en) * 2019-11-21 2023-05-12 圆通速递有限公司 A method and system for face-to-face information positioning based on deep learning
CN110909797B (en) * 2019-11-22 2023-05-05 北京深睿博联科技有限责任公司 Image detection method and device, equipment and storage medium
CN110705540B (en) * 2019-11-25 2024-05-31 中国农业科学院农业信息研究所 Image recognition method and device for pointer-type instrument in veterinary drug production based on RFID and deep learning
CN111105393B (en) * 2019-11-25 2023-04-18 长安大学 Grape disease and pest identification method and device based on deep learning
CN111144209B (en) * 2019-11-25 2024-07-02 浙江工商大学 Monitoring video head detection method based on heterogeneous multi-branch deep convolutional neural network
CN110956115B (en) * 2019-11-26 2023-09-29 证通股份有限公司 Scene recognition method and device
CN112949814B (en) * 2019-11-26 2024-04-26 联合汽车电子有限公司 Compression and acceleration method and device of convolutional neural network and embedded device
CN111046928B (en) * 2019-11-27 2023-05-23 上海交通大学 Single-stage real-time universal target detector and method with accurate positioning
CN111062278B (en) * 2019-12-03 2023-04-07 西安工程大学 Abnormal behavior identification method based on improved residual error network
CN111145195B (en) * 2019-12-03 2023-02-24 上海海事大学 A Method for Contour Detection of Portraits in Video Based on Lightweight Deep Neural Networks
CN110986949B (en) * 2019-12-04 2023-05-09 日照职业技术学院 A Path Recognition Method Based on Artificial Intelligence Platform
CN110942144B (en) * 2019-12-05 2023-05-02 深圳牛图科技有限公司 Neural network construction method integrating automatic training, checking and reconstruction
CN111027449B (en) * 2019-12-05 2023-05-30 光典信息发展有限公司 Positioning and identifying method for paper archive electronic image archive chapter
CN111178148B (en) * 2019-12-06 2023-06-02 天津大学 A ground target geographic coordinate positioning method based on UAV vision system
CN110992238B (en) * 2019-12-06 2023-10-17 上海电力大学 A blind detection method for digital image tampering based on dual-channel network
CN111008603B (en) * 2019-12-08 2023-04-18 中南大学 Multi-class target rapid detection method for large-scale remote sensing image
CN111160115B (en) * 2019-12-10 2023-05-02 上海工程技术大学 Video pedestrian re-identification method based on twin double-flow 3D convolutional neural network
CN111161217B (en) * 2019-12-10 2023-04-18 中国民航大学 Conv-LSTM multi-scale feature fusion-based fuzzy detection method
CN111179338B (en) * 2019-12-10 2023-08-04 同济大学 Lightweight target positioning method for mobile power supply receiving end
CN111062297B (en) * 2019-12-11 2023-05-23 青岛科技大学 Violent abnormal behavior detection method based on EANN deep learning model
CN111079642B (en) * 2019-12-13 2023-11-14 国网浙江余姚市供电有限公司 Line removable monitoring method and device and computer-readable medium
CN110956157A (en) * 2019-12-14 2020-04-03 深圳先进技术研究院 Deep learning remote sensing image target detection method and device based on candidate frame selection
CN111178178B (en) * 2019-12-16 2023-10-10 汇纳科技股份有限公司 Multi-scale pedestrian re-identification method, system, medium and terminal combined with regional distribution
CN111091101B (en) * 2019-12-23 2023-06-02 中国科学院自动化研究所 High-precision pedestrian detection method, system and device based on one-step method
CN111126310B (en) * 2019-12-26 2023-03-24 华侨大学 Pedestrian gender identification method based on scene migration
CN111178251B (en) * 2019-12-27 2023-07-28 汇纳科技股份有限公司 Pedestrian attribute identification method and system, storage medium and terminal
CN111160527B (en) 2019-12-27 2025-11-14 歌尔股份有限公司 A target recognition method and apparatus based on the MASK RCNN network model
CN111161295B (en) * 2019-12-30 2023-11-21 神思电子技术股份有限公司 Dish image background stripping method
CN111176820B (en) * 2019-12-31 2021-06-25 中科院计算技术研究所大数据研究院 A method and device for allocating edge computing tasks based on deep neural network
CN111160274B (en) * 2019-12-31 2023-03-24 合肥湛达智能科技有限公司 Pedestrian detection method based on binaryzation fast RCNN (radar cross-correlation neural network)
CN111199212B (en) * 2020-01-02 2023-04-07 西安工程大学 Pedestrian attribute identification method based on attention model
CN111209952B (en) * 2020-01-03 2023-05-30 西安工业大学 Underwater target detection method based on improved SSD and migration learning
CN111209860B (en) * 2020-01-06 2023-04-07 上海海事大学 Video attendance system and method based on deep learning and reinforcement learning
CN111259898B (en) * 2020-01-08 2023-03-24 西安电子科技大学 Crop segmentation method based on unmanned aerial vehicle aerial image
CN111259736B (en) * 2020-01-08 2023-04-07 上海海事大学 Real-time pedestrian detection method based on deep learning in complex environment
CN111275711B (en) * 2020-01-08 2023-04-07 西安电子科技大学 Real-time image semantic segmentation method based on lightweight convolutional neural network model
CN111242010A (en) * 2020-01-10 2020-06-05 厦门博海中天信息科技有限公司 Method for judging and identifying identity of litter worker based on edge AI
CN111260658B (en) * 2020-01-10 2023-10-17 厦门大学 A deep reinforcement learning method for image segmentation
CN111242839B (en) * 2020-01-13 2023-04-21 华南理工大学 Image scaling and clipping method based on scale level
CN111242127B (en) * 2020-01-15 2023-02-24 上海应用技术大学 Vehicle detection method with granularity level multi-scale characteristic based on asymmetric convolution
CN111209887B (en) * 2020-01-15 2023-04-07 西安电子科技大学 SSD model optimization method for small target detection
CN113128316B (en) * 2020-01-15 2024-08-02 北京四维图新科技股份有限公司 Target detection method and device
CN111259800A (en) * 2020-01-16 2020-06-09 天津大学 A detection method of unmanned special vehicle based on neural network
CN111291785B (en) * 2020-01-16 2024-11-19 中国平安人寿保险股份有限公司 Target detection method, device, equipment and storage medium
CN111222519B (en) * 2020-01-16 2023-03-24 西北大学 Construction method, method and device of hierarchical colored drawing manuscript line extraction model
CN111275688B (en) * 2020-01-19 2023-12-12 合肥工业大学 Small target detection method based on context feature fusion and screening based on attention mechanism
CN111275171B (en) * 2020-01-19 2023-07-04 合肥工业大学 A small target detection method based on multi-scale super-resolution reconstruction based on parameter sharing
CN111199220B (en) * 2020-01-21 2023-04-28 北方民族大学 A Lightweight Deep Neural Network Method for People Detection and People Counting in Elevators
CN111277751B (en) * 2020-01-22 2021-06-15 Oppo广东移动通信有限公司 Photographing method and device, storage medium and electronic equipment
CN111292366B (en) * 2020-02-17 2023-03-10 华侨大学 Visual driving ranging algorithm based on deep learning and edge calculation
CN111339871B (en) * 2020-02-18 2022-09-16 中国电子科技集团公司第二十八研究所 A method and device for judging the distribution pattern of target groups based on convolutional neural network
CN111291820B (en) * 2020-02-19 2023-05-30 中国电子科技集团公司第二十八研究所 A Target Detection Method Combining Location Information and Classification Information
CN111428751B (en) * 2020-02-24 2022-12-23 清华大学 Object detection method based on compressed sensing and convolutional network
CN111428567B (en) * 2020-02-26 2024-02-02 沈阳大学 A pedestrian tracking system and method based on affine multi-task regression
CN111368673B (en) * 2020-02-26 2023-04-07 华南理工大学 Method for quickly extracting human body key points based on neural network
CN113324864B (en) * 2020-02-28 2022-09-20 南京理工大学 Pantograph carbon slide plate abrasion detection method based on deep learning target detection
CN111339967B (en) * 2020-02-28 2023-04-07 长安大学 Pedestrian detection method based on multi-view graph convolution network
CN111339975B (en) * 2020-03-03 2023-04-21 华东理工大学 Object Detection, Recognition and Tracking Method Based on Central Scale Prediction and Siamese Neural Network
CN111368726B (en) * 2020-03-04 2023-11-10 西安咏圣达电子科技有限公司 Construction site operation face personnel number statistics method, system, storage medium and device
CN111428586B (en) * 2020-03-09 2023-05-16 同济大学 3D Human Pose Estimation Method Based on Feature Fusion and Sample Enhancement
CN111429410B (en) * 2020-03-13 2023-09-01 杭州电子科技大学 A system and method for object X-ray image material discrimination based on deep learning
CN111461291B (en) * 2020-03-13 2023-05-12 西安科技大学 Long-distance pipeline inspection method based on YOLOv3 pruning network and deep learning dehazing model
CN111414909B (en) * 2020-03-16 2023-05-12 上海富瀚微电子股份有限公司 Target detection method and device
CN111460924B (en) * 2020-03-16 2023-04-07 上海师范大学 Gate ticket-evading behavior detection method based on target detection
CN111368453B (en) * 2020-03-17 2023-07-07 创新奇智(合肥)科技有限公司 Fabric cutting optimization method based on deep reinforcement learning
CN111753625B (en) * 2020-03-18 2024-04-09 北京沃东天骏信息技术有限公司 Pedestrian detection method, device, equipment and medium
CN111462132A (en) * 2020-03-20 2020-07-28 西北大学 A method and system for video object segmentation based on deep learning
CN111488805B (en) * 2020-03-24 2023-04-25 广州大学 A Video Behavior Recognition Method Based on Salient Feature Extraction
CN111563525A (en) * 2020-03-25 2020-08-21 北京航空航天大学 Moving target detection method based on YOLOv3-Tiny
CN111414997B (en) * 2020-03-27 2023-06-06 中国人民解放军空军工程大学 A Method for Battlefield Target Recognition Based on Artificial Intelligence
CN111310861B (en) * 2020-03-27 2023-05-23 西安电子科技大学 A license plate recognition and location method based on deep neural network
CN111310773B (en) * 2020-03-27 2023-03-24 西安电子科技大学 Efficient license plate positioning method of convolutional neural network
CN111460980B (en) * 2020-03-30 2023-04-07 西安工程大学 Multi-scale detection method for small-target pedestrian based on multi-semantic feature fusion
CN111462085B (en) * 2020-03-31 2023-09-19 上海大学 Digital image local filtering forensics method based on convolutional neural network
CN111553199A (en) * 2020-04-07 2020-08-18 厦门大学 An automatic detection technology for motor vehicle traffic violations based on computer vision
CN111462108B (en) * 2020-04-13 2023-05-02 山西新华防化装备研究院有限公司 Machine learning-based head-face product design ergonomics evaluation operation method
CN111523645B (en) * 2020-04-16 2023-04-18 北京航天自动控制研究所 Convolutional neural network design method for improving small-scale target detection and identification performance
CN111597900B (en) * 2020-04-16 2023-10-24 浙江工业大学 How to identify illegal dog walking
CN111597897B (en) * 2020-04-16 2023-10-24 浙江工业大学 Parking space identification method in expressway service area
CN111695403B (en) * 2020-04-19 2024-03-22 东风汽车股份有限公司 Depth perception convolutional neural network-based 2D and 3D image synchronous detection method
CN111476314B (en) * 2020-04-27 2023-03-07 中国科学院合肥物质科学研究院 Fuzzy video detection method integrating optical flow algorithm and deep learning
CN111563440A (en) * 2020-04-29 2020-08-21 上海海事大学 A target detection method based on heterogeneous convolution with multi-kernel iterative RPN
CN111652846B (en) * 2020-04-30 2022-08-16 成都数之联科技股份有限公司 Semiconductor defect identification method based on characteristic pyramid convolution neural network
CN111597939B (en) * 2020-05-07 2023-04-18 西安电子科技大学 High-speed rail line nest defect detection method based on deep learning
CN111783685B (en) * 2020-05-08 2025-01-21 西安建筑科技大学 An improved target detection algorithm based on a single-stage network model
CN111582452B (en) * 2020-05-09 2023-10-27 北京百度网讯科技有限公司 Methods and devices for generating neural network models
CN111597945B (en) * 2020-05-11 2023-08-18 济南博观智能科技有限公司 A target detection method, device, equipment and medium
CN111783934B (en) * 2020-05-15 2024-06-21 北京迈格威科技有限公司 Convolutional neural network construction method, device, equipment and medium
CN111598951B (en) * 2020-05-18 2022-09-30 清华大学 Method, device and storage medium for identifying space target
CN111783523B (en) * 2020-05-19 2022-10-21 中国人民解放军93114部队 A method for detecting rotating objects in remote sensing images
CN111709449B (en) * 2020-05-20 2023-08-18 西安理工大学 Multi-layer feature fusion small-scale target detection method based on clustering algorithm
CN112001878A (en) * 2020-05-21 2020-11-27 合肥合工安驰智能科技有限公司 Deep learning ore scale measuring method based on binarization neural network and application system
CN111881714B (en) * 2020-05-22 2023-11-21 北京交通大学 An unsupervised cross-domain person re-identification method
CN113793292B (en) * 2020-05-25 2025-09-16 阿里巴巴集团控股有限公司 Data processing method, device, electronic equipment and storage medium
CN111709313B (en) * 2020-05-27 2022-07-29 杭州电子科技大学 Pedestrian re-identification method based on local and channel combination characteristics
CN111709311B (en) * 2020-05-27 2023-11-28 西安理工大学 A pedestrian re-identification method based on multi-scale convolution feature fusion
CN111626196B (en) * 2020-05-27 2023-05-16 西南石油大学 Knowledge-graph-based intelligent analysis method for body structure of typical bovine animal
CN111832608B (en) * 2020-05-29 2023-09-12 上海海事大学 A multi-wear particle identification method in ferrogram images based on the single-stage detection model yolov3
CN111652216B (en) * 2020-06-03 2023-04-07 北京工商大学 Multi-scale target detection model method based on metric learning
CN111652930B (en) * 2020-06-04 2024-02-27 上海媒智科技有限公司 Image target detection method, system and equipment
CN111709336B (en) * 2020-06-08 2024-04-26 杭州像素元科技有限公司 A method, device and readable storage medium for detecting pedestrians on a highway
CN111881932B (en) * 2020-06-11 2023-09-15 中国人民解放军战略支援部队信息工程大学 A FasterRCNN target detection algorithm for military aircraft
CN111860587B (en) * 2020-06-12 2024-02-02 长安大学 A detection method for small targets in pictures
CN111738124B (en) * 2020-06-15 2023-08-22 西安电子科技大学 Cloud Detection Method of Remote Sensing Image Based on Gabor Transform and Attention
CN113807152A (en) * 2020-06-17 2021-12-17 阿里巴巴集团控股有限公司 Image processing and model training method and device
CN111709935B (en) * 2020-06-17 2023-04-07 西安科技大学 Real-time coal gangue positioning and identifying method for ground moving belt
CN111797836B (en) * 2020-06-18 2024-04-26 中国空间技术研究院 A deep learning-based obstacle segmentation method for extraterrestrial rover
CN111723743A (en) * 2020-06-19 2020-09-29 北京邮电大学 A Fast Pedestrian Detection Method at Small Scale
CN111832630A (en) * 2020-06-23 2020-10-27 成都恒创新星科技有限公司 Target detection method based on first-order gradient neural network
CN111784652B (en) * 2020-06-24 2024-02-06 西安电子科技大学 MRI (magnetic resonance imaging) segmentation method based on reinforcement learning multi-scale neural network
CN111814621B (en) * 2020-06-29 2024-01-23 中国科学院合肥物质科学研究院 A multi-scale vehicle and pedestrian detection method and device based on attention mechanism
CN111767847B (en) * 2020-06-29 2023-06-09 佛山市南海区广工大数控装备协同创新研究院 A Pedestrian Multi-Target Tracking Method Integrating Object Detection and Association
CN111832450B (en) * 2020-06-30 2023-11-28 成都睿沿科技有限公司 Knife holding detection method based on image recognition
CN111767878B (en) * 2020-07-03 2022-11-08 中国科学院自动化研究所 Deep learning-based traffic sign detection method and system in embedded device
CN112199983B (en) * 2020-07-08 2024-06-18 北京航空航天大学 Long-time large-range pedestrian re-identification method based on multi-level screening
CN111986145B (en) * 2020-07-09 2024-06-21 浙江工业大学 A bearing roller defect detection method based on Faster-RCNN
CN111860265B (en) * 2020-07-10 2024-01-05 武汉理工大学 Multi-detection-frame loss balanced road scene understanding algorithm based on sample loss
CN111667030B (en) * 2020-07-13 2023-04-07 华东理工大学 Method, system and storage medium for realizing remote sensing image target detection based on deep neural network
CN111832479B (en) * 2020-07-14 2023-08-01 西安电子科技大学 Video Object Detection Method Based on Improved Adaptive Anchor R-CNN
CN111986149A (en) * 2020-07-16 2020-11-24 江西斯源科技有限公司 A method for detecting plant diseases and insect pests based on convolutional neural network
CN111860637B (en) * 2020-07-17 2023-11-21 河南科技大学 A single-shot multi-frame infrared target detection method
CN111986126B (en) * 2020-07-17 2022-05-24 浙江工业大学 Multi-target detection method based on improved VGG16 network
CN111832513B (en) * 2020-07-21 2024-02-09 西安电子科技大学 Real-time football target detection method based on neural network
CN111881803B (en) * 2020-07-22 2023-10-31 安徽农业大学 An animal face recognition method based on improved YOLOv3
CN112036437B (en) * 2020-07-28 2024-06-07 农业农村部南京农业机械化研究所 Rice seedling detection model based on improved YOLOV network and method thereof
CN112001259A (en) * 2020-07-28 2020-11-27 联芯智能(南京)科技有限公司 Aerial weak human body target intelligent detection method based on visible light image
CN111860508B (en) * 2020-07-28 2024-07-02 平安科技（深圳）有限公司 Image sample selection method and related equipment
CN111915583B (en) * 2020-07-29 2024-02-09 西安电子科技大学 Vehicle and pedestrian detection method based on vehicle-mounted thermal infrared imager in complex scene
CN111985365A (en) * 2020-08-06 2020-11-24 合肥学院 A straw burning monitoring method and system based on target detection technology
CN112115291B (en) * 2020-08-12 2024-02-27 南京止善智能科技研究院有限公司 A three-dimensional indoor model retrieval method based on deep learning
CN111985464B (en) * 2020-08-13 2023-08-22 山东大学 Court judgment document-oriented multi-scale learning text recognition method and system
CN111986172B (en) * 2020-08-18 2024-06-04 华北电力科学研究院有限责任公司 Infrared image fault detection method and device for power equipment
CN111984879A (en) * 2020-08-19 2020-11-24 交控科技股份有限公司 User guidance method, device, device and storage medium applied to trains
CN111985473A (en) * 2020-08-20 2020-11-24 中再云图技术有限公司 A method of out-of-store business identification
CN112001385B (en) * 2020-08-20 2024-02-06 长安大学 Target cross-domain detection and understanding method, system, equipment and storage medium
CN111986186B (en) * 2020-08-25 2024-03-22 华中科技大学 Quantitative in-furnace PCB patch defect online detection system and method
CN112001339B (en) * 2020-08-27 2024-02-23 杭州电子科技大学 A real-time monitoring method of pedestrian social distance based on YOLO v4
CN112364974B (en) * 2020-08-28 2024-02-09 西安电子科技大学 YOLOv3 algorithm based on activation function improvement
CN112149664B (en) * 2020-09-04 2024-05-07 浙江工业大学 Target detection method for optimizing classification and positioning tasks
CN112101434B (en) * 2020-09-04 2022-09-09 河南大学 Infrared image weak and small target detection method based on improved YOLO v3
CN112464765B (en) * 2020-09-10 2022-09-23 天津师范大学 A safety helmet detection method based on single-pixel feature amplification and its application
CN115661491A (en) * 2020-09-15 2023-01-31 重庆市农业科学院 Monitoring method for pest control in tea tree planting
CN112347843B (en) * 2020-09-18 2024-10-18 深圳数联天下智能科技有限公司 Method and related device for training wrinkle detection model
CN112163492B (en) * 2020-09-21 2023-09-08 华南理工大学 A long-term cross-scenario optimized traffic object detection method, system and medium
CN112115885B (en) * 2020-09-22 2023-08-11 中国农业科学院农业信息研究所 Fruit tree fruiting branch shearing point positioning method based on deep convolutional neural network
CN112215100B (en) * 2020-09-27 2024-02-09 浙江工业大学 Target detection method for degraded image under unbalanced training sample
CN112347851B (en) * 2020-09-30 2023-02-21 山东理工大学 Construction method of multi-target detection network, multi-target detection method and device
CN112085126B (en) * 2020-09-30 2023-12-12 浙江大学 A single-sample target detection method focusing on classification tasks
CN112200045B (en) * 2020-09-30 2024-03-19 华中科技大学 Method and application of remote sensing image target detection model based on context enhancement
CN112183430B (en) * 2020-10-12 2024-04-05 河北工业大学 Sign language recognition method and device based on dual neural network
CN112232411B (en) * 2020-10-15 2024-05-14 苏州凌图科技有限公司 HarDNet-Lite optimization method in embedded platform
CN112257796B (en) * 2020-10-28 2024-06-28 辽宁工程技术大学 An image integration method based on convolutional neural network with selective feature connection
CN112419237B (en) * 2020-11-03 2023-06-30 中国计量大学 Deep learning-based automobile clutch master cylinder groove surface defect detection method
CN112381792B (en) * 2020-11-13 2023-05-23 中国人民解放军空军工程大学 Intelligent imaging on-line detection method for radar wave-absorbing coating/electromagnetic shielding film damage based on deep learning
CN112446308B (en) * 2020-11-16 2024-09-13 北京科技大学 Pedestrian detection method based on semantic enhancement multi-scale feature pyramid fusion
CN112396000B (en) * 2020-11-19 2023-09-05 中山大学 A Construction Method of Depth Information Transmission Model for Multimodal Dense Prediction
CN112308062B (en) * 2020-11-23 2022-08-23 浙江卡易智慧医疗科技有限公司 Medical image access number identification method in complex background image
CN112434828B (en) * 2020-11-23 2023-05-16 南京富岛软件有限公司 Intelligent safety protection identification method in 5T operation and maintenance
CN112580778A (en) * 2020-11-25 2021-03-30 江苏集萃未来城市应用技术研究所有限公司 Job worker mobile phone use detection method based on YOLOv5 and Pose-animation
CN112348036B (en) * 2020-11-26 2025-01-14 北京工业大学 Adaptive object detection method based on lightweight residual learning and deconvolution cascade
CN112487979B (en) * 2020-11-30 2023-08-04 北京百度网讯科技有限公司 Target detection method and model training method, device, electronic device and medium
CN112528826B (en) * 2020-12-04 2024-02-02 江苏省农业科学院 Control method of picking device based on 3D visual perception
CN112633086B (en) * 2020-12-09 2024-01-26 西安电子科技大学 Near-infrared pedestrian monitoring methods, systems, media, and equipment based on multi-task EfficientDet
CN112396036B (en) * 2020-12-09 2023-08-08 中山大学 An Occluded Person Re-Identification Method Combining Spatial Transformation Network and Multi-Scale Feature Extraction
CN112770325B (en) * 2020-12-09 2022-12-16 华南理工大学 Cognitive internet of vehicles spectrum sensing method based on deep learning
CN112560627A (en) * 2020-12-09 2021-03-26 江苏集萃未来城市应用技术研究所有限公司 Real-time detection method for abnormal behaviors of construction site personnel based on neural network
CN112613359B (en) * 2020-12-09 2024-02-02 苏州玖合智能科技有限公司 Construction method of neural network for detecting abnormal behaviors of personnel
CN112382388A (en) * 2020-12-14 2021-02-19 中南大学 Early warning method for adverse pressure sore event
CN112560682A (en) * 2020-12-16 2021-03-26 重庆守愚科技有限公司 Valve automatic detection method based on deep learning
CN112465815B (en) * 2020-12-17 2023-09-19 杭州电子科技大学 A remote sensing target saliency detection method based on edge-subject fusion information
CN112651441B (en) * 2020-12-25 2022-08-16 深圳市信义科技有限公司 Fine-grained non-motor vehicle feature detection method, storage medium and computer equipment
CN112634367A (en) * 2020-12-25 2021-04-09 天津大学 Anti-occlusion object pose estimation method based on deep neural network
CN112699808B (en) * 2020-12-31 2024-06-07 深圳市华尊科技股份有限公司 Dense target detection method, electronic equipment and related products
CN112613472B (en) * 2020-12-31 2022-04-26 上海交通大学 A pedestrian detection method and system based on deep search matching
CN112733848B (en) * 2021-01-08 2022-11-04 中国电子科技集团公司第二十八研究所 Object detection method based on multi-scale features and dilated inverse residual full connection
CN112733714B (en) * 2021-01-11 2024-03-01 北京大学 VGG network-based automatic crowd counting image recognition method
CN112784921A (en) * 2021-02-02 2021-05-11 西北工业大学 Task attention guided small sample image complementary learning classification algorithm
CN112556682B (en) * 2021-02-07 2023-06-23 天津蓝鳍海洋工程有限公司 Automatic detection algorithm for underwater composite sensor target
CN112700444B (en) * 2021-02-19 2023-06-23 中国铁道科学研究院集团有限公司铁道建筑研究所 Bridge bolt detection method based on self-attention and center point regression model
CN112862796B (en) * 2021-02-23 2024-11-29 中国农业机械化科学研究院 Multi-category external quality detection method and detection device for kernel and fruit
CN112949508B (en) * 2021-03-08 2024-07-19 咪咕文化科技有限公司 Model training method, pedestrian detection method, electronic device and readable storage medium
CN112906718B (en) * 2021-03-09 2023-08-22 西安电子科技大学 A multi-target detection method based on convolutional neural network
CN113012208B (en) * 2021-03-22 2024-05-17 上海应用技术大学 Multi-view remote sensing image registration method and system
CN112906658B (en) * 2021-03-30 2025-01-10 航天时代飞鸿技术有限公司 A lightweight automatic detection method for UAV reconnaissance of ground targets
CN113312961A (en) * 2021-04-03 2021-08-27 国家计算机网络与信息安全管理中心 Logo recognition acceleration method
CN113221957B (en) * 2021-04-17 2024-04-16 南京航空航天大学 A radar information fusion feature enhancement method based on Centernet
CN113112511B (en) * 2021-04-19 2024-01-05 新东方教育科技集团有限公司 Method and device for correcting test paper, storage medium and electronic equipment
CN113076957A (en) * 2021-04-21 2021-07-06 河南大学 RGB-D image saliency target detection method based on cross-modal feature fusion
CN113011398A (en) * 2021-04-28 2021-06-22 北京邮电大学 Target change detection method and device for multi-temporal remote sensing image
CN113177545B (en) * 2021-04-29 2023-08-04 北京百度网讯科技有限公司 Target object detection method, target object detection device, electronic equipment and storage medium
CN113158968A (en) * 2021-05-10 2021-07-23 苏州大学 Embedded object cognitive system based on image processing
CN113408340B (en) * 2021-05-12 2024-03-29 北京化工大学 Dual-polarization SAR small ship detection method based on enhanced feature pyramid
CN113312995B (en) * 2021-05-18 2023-02-14 华南理工大学 Anchor-free vehicle-mounted pedestrian detection method based on central axis
CN113221787B (en) * 2021-05-18 2023-09-29 西安电子科技大学 Pedestrian multi-target tracking method based on multi-difference fusion
CN113297961B (en) * 2021-05-24 2023-11-17 南京邮电大学 A target tracking method based on boundary feature fusion twin recurrent neural networks
CN113222064A (en) * 2021-05-31 2021-08-06 苏州晗林信息技术发展有限公司 Image target object real-time detection method, system, terminal and storage medium
CN113343853B (en) * 2021-06-08 2024-06-14 深圳格瑞健康科技有限公司 Intelligent screening method and device for dental caries of children
CN113379699A (en) * 2021-06-08 2021-09-10 上海电机学院 Transmission line insulator defect detection method based on deep learning
CN113379709B (en) * 2021-06-16 2024-03-08 浙江工业大学 A three-dimensional target detection method based on sparse multi-scale voxel feature fusion
CN113449634A (en) * 2021-06-28 2021-09-28 上海翰声信息技术有限公司 Video detection method and device for processing under strong light environment
CN113379718B (en) * 2021-06-28 2024-02-02 北京百度网讯科技有限公司 A target detection method, device, electronic equipment and readable storage medium
CN113469254B (en) * 2021-07-02 2024-04-16 上海应用技术大学 Target detection method and system based on target detection model
CN113486810B (en) * 2021-07-08 2024-06-18 国网江苏省电力有限公司徐州供电分公司 An intelligent identification method for poached birds in parks
CN113627257B (en) * 2021-07-09 2024-09-10 上海智臻智能网络科技股份有限公司 Detection method, detection system, device and storage medium
CN113449743B (en) * 2021-07-12 2022-12-09 西安科技大学 Coal dust particle feature extraction method
CN113642410B (en) * 2021-07-15 2024-03-29 南京航空航天大学 Method for detecting ampullaria gigas eggs based on multi-scale feature fusion and dynamic convolution
CN113361491A (en) * 2021-07-19 2021-09-07 厦门大学 Method for predicting pedestrian crossing intention of unmanned automobile
CN113657174A (en) * 2021-07-21 2021-11-16 北京中科慧眼科技有限公司 Vehicle pseudo-3D information detection method and device and automatic driving system
CN113487600B (en) * 2021-07-27 2024-05-03 大连海事大学 Feature enhancement scale self-adaptive perception ship detection method
CN113592825A (en) * 2021-08-02 2021-11-02 安徽理工大学 YOLO algorithm-based real-time coal gangue detection method
CN113591735A (en) * 2021-08-04 2021-11-02 上海新纪元机器人有限公司 Pedestrian detection method and system based on deep learning
CN113591854B (en) * 2021-08-12 2023-09-26 中国海洋大学 A low-redundancy and fast reconstruction method for plankton holograms
CN113805151B (en) * 2021-08-17 2024-09-10 青岛本原微电子有限公司 Medium-frequency radar target detection method based on attention mechanism
US12462575B2 (en) 2021-08-19 2025-11-04 Tesla, Inc. Vision-based machine learning model for autonomous driving with adjustable virtual camera
CN113869361A (en) * 2021-08-20 2021-12-31 深延科技(北京)有限公司 Model training method, target detection method and related device
CN113706491B (en) * 2021-08-20 2024-02-13 西安电子科技大学 Meniscal injury grading method based on hybrid attention weakly supervised transfer learning
CN113989630B (en) * 2021-08-31 2024-04-23 中通服公众信息产业股份有限公司 A lens occlusion identification method based on semantic analysis
CN113822185B (en) * 2021-09-09 2024-10-29 安徽农业大学 Method for detecting daily behaviors of group-raised pigs
CN113887330A (en) * 2021-09-10 2022-01-04 国网吉林省电力有限公司 Target detection system based on remote sensing image
CN113780193B (en) * 2021-09-15 2024-09-24 易采天成(郑州)信息技术有限公司 RCNN-based cattle group target detection method and RCNN-based cattle group target detection equipment
CN113962933A (en) * 2021-09-15 2022-01-21 上海大学 PCB defect image detection method based on improved YOLOv3
CN113807243B (en) * 2021-09-16 2023-12-05 上海交通大学 Water obstacle detection system and method based on attention to unknown target
CN113887341B (en) * 2021-09-16 2025-04-29 同济大学 A method for human skeleton action recognition based on parallel convolutional neural network
CN113762209B (en) * 2021-09-22 2025-03-18 重庆邮电大学 A multi-scale parallel feature fusion landmark detection method based on YOLO
CN114067186B (en) * 2021-09-26 2024-04-16 北京建筑大学 Pedestrian detection method and device, electronic equipment and storage medium
CN113902024B (en) * 2021-10-20 2024-06-04 浙江大立科技股份有限公司 Small-volume target detection and identification method based on deep learning and dual-band fusion
CN114091518A (en) * 2021-10-21 2022-02-25 安徽深核信息技术有限公司 A Multi-stage Weak Object Image Detection Method Using Combined Features
CN113901944B (en) * 2021-10-25 2024-04-09 大连理工大学 Marine organism target detection method based on improved YOLO algorithm
CN114067228B (en) * 2021-10-26 2024-11-29 神思电子技术股份有限公司 Target detection method and system for enhancing foreground and background distinction
CN115082909B (en) * 2021-11-03 2024-04-12 中国人民解放军陆军军医大学第一附属医院 Method and system for identifying lung lesions
CN113989518B (en) * 2021-11-08 2024-11-15 中国科学院合肥物质科学研究院 A lightweight target detection method for intelligent terminals
CN114170531B (en) * 2021-11-23 2024-08-09 北京航天自动控制研究所 Infrared image target detection method and device based on difficult sample transfer learning
CN114170633B (en) * 2021-12-06 2024-08-02 南开大学 Sea surface small pedestrian detection method based on collaborative supervision
CN114241220B (en) * 2021-12-10 2025-08-15 北京科技大学 Unmanned aerial vehicle ground object detection method
CN114399644B (en) * 2021-12-15 2025-02-07 北京邮电大学 Small sample target detection method and device
CN114419663A (en) * 2021-12-17 2022-04-29 江西洪都航空工业集团有限责任公司 Detection method for mask worn by human body
CN114359644B (en) * 2021-12-22 2024-04-16 华南农业大学 Crop pest and disease recognition method based on improved VGG-16 network
CN114283320B (en) * 2021-12-25 2024-06-14 福州大学 A branchless structured target detection method based on full convolution
CN114332008B (en) * 2021-12-28 2024-06-28 福州大学 Unsupervised defect detection and positioning method based on multi-level feature reconstruction
CN114332029A (en) * 2021-12-30 2022-04-12 上海华力微电子有限公司 Method and device for classifying abnormal crystal back images by using neural network model
CN114372971A (en) * 2022-01-10 2022-04-19 重庆邮电大学 Rapid three-dimensional target detection algorithm based on binocular vision
CN114359838B (en) * 2022-01-14 2025-03-28 北京理工大学重庆创新中心 A cross-modal pedestrian detection method based on Gaussian cross-attention network
CN114495166A (en) * 2022-01-17 2022-05-13 北京小龙潜行科技有限公司 A kind of pasture changing shoe action recognition method applied to edge computing equipment
CN114419589B (en) * 2022-01-17 2025-04-25 东南大学 A road object detection method based on attention feature enhancement module
CN114418000B (en) * 2022-01-20 2024-11-19 辽宁工程技术大学 A remote sensing image intelligent interpretation method and system based on adaptive perceptron
CN114565753B (en) * 2022-02-22 2025-04-15 电子科技大学长三角研究院(湖州) A method for identifying small targets of drones based on improved YOLOv4 network
CN114612769B (en) * 2022-03-14 2023-05-26 电子科技大学 Integrated sensing infrared imaging ship detection method integrated with local structure information
CN114638971B (en) * 2022-03-21 2025-01-07 天津大学 Object detection method based on adaptive fusion of multi-level local and global features
CN114884775A (en) * 2022-03-31 2022-08-09 南京邮电大学 Deep learning-based large-scale MIMO system channel estimation method
CN114863097B (en) * 2022-04-06 2024-05-31 北京航空航天大学 A method for infrared dim small target detection based on attention mechanism convolutional neural network
CN114821477B (en) * 2022-05-05 2025-06-27 南京大学 A video recognition method for multi-person clothing features suitable for complex scenes
CN115019036B (en) * 2022-05-10 2024-02-27 西北工业大学 A small-sample semantic segmentation method for learning non-target knowledge
CN114998616A (en) * 2022-05-13 2022-09-02 南京林业大学 Multi-scale target number statistics network based on online mask perception
CN114943986B (en) * 2022-05-31 2024-09-27 武汉理工大学 Regional pedestrian detection lighting method and system based on camera image segmentation
CN115082386B (en) * 2022-06-07 2024-04-26 华南理工大学 Injection molded parts defect detection method, device and medium based on normal sample auxiliary feature extraction
CN115272175B (en) * 2022-06-16 2025-12-02 上海互觉科技有限公司 A Surface Defect Detection Method and System Based on Multi-Light Source Collaboration
CN115063701B (en) * 2022-06-21 2025-06-13 南京理工大学 A small target detection method for UAV aerial photography based on improved YOLOv4
CN115147711B (en) * 2022-07-23 2024-07-16 河南大学 Underwater target detection network and method based on improvement RETINANET
CN115205686B (en) * 2022-07-26 2025-10-24 中山大学·深圳 A SAR image aircraft detection method and device
CN115457287A (en) * 2022-08-30 2022-12-09 中国电子科技集团公司第三十八研究所 Method for simultaneously carrying out target detection and regression analysis by using convolutional neural network
CN115471498B (en) * 2022-10-10 2025-08-01 温州市华炜鞋材科技有限公司 Multi-angle waterproof monitoring shoe making machine for rain shoes production and method thereof
CN115512326B (en) * 2022-10-18 2025-07-25 上海寻序人工智能科技有限公司 BEV visual perception method based on multiple cameras
CN115601790A (en) * 2022-10-26 2023-01-13 功夫链（上海）体育文化发展有限公司（Cn） A method and system for close-range human detection based on super large convolution kernel
CN115546784B (en) * 2022-10-27 2025-12-19 奥特酷智能科技(南京)有限公司 3D target detection method based on deep learning
CN115423810B (en) * 2022-11-04 2023-03-14 国网江西省电力有限公司电力科学研究院 A Method for Analyzing the Icing Form of Wind Turbine Blades
CN116403151A (en) * 2022-12-20 2023-07-07 南京工业大学 A subway crowd density estimation method based on adaptive deep neural network
CN116468928B (en) * 2022-12-29 2023-12-19 长春理工大学 Thermal infrared small target detection method based on visual perception correlator
CN116052214B (en) * 2023-01-18 2025-06-27 天津大学 Pedestrian searching method based on dynamic RoI feature extraction
CN116168270B (en) * 2023-02-21 2026-01-02 安徽理工大学 Lightweight Coal Gangue Detection Model and Method Based on Parallel Deep Residual Network
CN117079342A (en) * 2023-02-27 2023-11-17 盛视科技股份有限公司 Man-box interaction detection method and system based on two-stage target detection
CN116523830A (en) * 2023-03-10 2023-08-01 上海交通大学 3D target detection method, system, medium and terminal
CN117197687B (en) * 2023-03-13 2025-11-18 西南科技大学 A detection method for dense small targets in UAV aerial photography
CN116524517B (en) * 2023-03-31 2025-09-30 西安电子科技大学 Electricity meter image target recognition method based on deep learning
CN116524293B (en) * 2023-04-10 2024-01-30 哈尔滨市科佳通用机电股份有限公司 Brake adjuster pull rod head loss fault identification method and system based on deep learning
CN116433979A (en) * 2023-04-18 2023-07-14 安徽理工大学 Tunnel anomaly detection method based on small feature-aware pyramid network
CN116563734B (en) * 2023-05-12 2025-11-21 贵州理工学院 Real-time target detection method and system for unmanned aerial vehicle platform
CN117237614B (en) * 2023-11-10 2024-02-06 江西啄木蜂科技有限公司 A method for detecting small targets floating on lake surface based on deep learning
CN118379848B (en) * 2024-06-21 2024-08-20 珠海华熠电子有限公司 Personnel safety monitoring system and method based on binocular camera

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105787439A (en) * 2016-02-04 2016-07-20 广州新节奏智能科技有限公司 Depth image human body joint positioning method based on convolution nerve network
CN105956608A (en) * 2016-04-21 2016-09-21 恩泊泰（天津）科技有限公司 Objective positioning and classifying algorithm based on deep learning
CN106934346A (en) * 2017-01-24 2017-07-07 北京大学 A kind of method of target detection performance optimization

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107341517B (en) * 2017-07-07 2020-08-11 哈尔滨工业大学 Multi-scale small object detection method based on deep learning inter-level feature fusion
CN107563349A (en) * 2017-09-21 2018-01-09 电子科技大学 A kind of Population size estimation method based on VGGNet

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105787439A (en) * 2016-02-04 2016-07-20 广州新节奏智能科技有限公司 Depth image human body joint positioning method based on convolution nerve network
CN105956608A (en) * 2016-04-21 2016-09-21 恩泊泰（天津）科技有限公司 Objective positioning and classifying algorithm based on deep learning
CN106934346A (en) * 2017-01-24 2017-07-07 北京大学 A kind of method of target detection performance optimization

Also Published As

Publication number Publication date
WO2019144575A1 (en) 2019-08-01
CN108399362A (en) 2018-08-14

Similar Documents

Publication Publication Date Title
CN108399362B (en) Rapid pedestrian detection method and device
CN113065558B (en) Lightweight small target detection method combined with attention mechanism
CN108830285B (en) Target detection method for reinforcement learning based on fast-RCNN
CN109993082B (en) Convolutional neural network road scene classification and road segmentation method
CN111160249B (en) Multi-target detection method based on cross-scale feature fusion in optical remote sensing images
CN110287826B (en) Video target detection method based on attention mechanism
CN112884742B (en) A multi-target real-time detection, recognition and tracking method based on multi-algorithm fusion
CN110472627A (en) One kind SAR image recognition methods end to end, device and storage medium
CN113052185A (en) Small sample target detection method based on fast R-CNN
JP6397379B2 (en) CHANGE AREA DETECTION DEVICE, METHOD, AND PROGRAM
WO2023116632A1 (en) Video instance segmentation method and apparatus based on spatio-temporal memory information
CN107341488A (en) A kind of SAR image target detection identifies integral method
CN115035295A (en) A Semantic Segmentation Method of Remote Sensing Image Based on Shared Convolution Kernel and Boundary Loss Function
CN111160407A (en) Deep learning target detection method and system
US11367206B2 (en) Edge-guided ranking loss for monocular depth prediction
CN113704276B (en) Map updating method, device, electronic device and computer-readable storage medium
CN118608781B (en) Semantic segmentation method, semantic segmentation device, electronic equipment and storage medium
CN111126278A (en) A method for optimizing and accelerating object detection model for few-category scenes
CN109165658B (en) A strong negative sample underwater target detection method based on Faster-RCNN
CN109635667A (en) A kind of vehicle detecting system based on Guided Faster-RCNN
CN118097365A (en) A lightweight object detection network design method for indoor semantic SLAM system
Aldhaheri et al. MACC Net: Multi-task attention crowd counting network
CN114842350B (en) Method and device for target detection in remote sensing images
CN110245660B (en) Webpage glance path prediction method based on saliency feature fusion
CN115170662A (en) Multi-target positioning method based on yolov3 and convolutional neural network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant