CN109145715B - Air-based pedestrian boundary-crossing detection method, device and system for rail transit - Google Patents
Air-based pedestrian boundary-crossing detection method, device and system for rail transit Download PDFInfo
- Publication number
- CN109145715B CN109145715B CN201810710554.9A CN201810710554A CN109145715B CN 109145715 B CN109145715 B CN 109145715B CN 201810710554 A CN201810710554 A CN 201810710554A CN 109145715 B CN109145715 B CN 109145715B
- Authority
- CN
- China
- Prior art keywords
- image
- space
- orbit
- boundary
- distance
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
Abstract
The invention provides a method, a device and a system for detecting an air-based pedestrian boundary intrusion of rail transit, wherein an anti-boundary intrusion area image is determined in an air-based rail image according to the proportion of the pixel distance to the real distance in the air-based rail image, and the air-based rail image is an image vertically shot downwards along a rail by an air-based image acquisition device; cutting the anti-invasion boundary area image into at least 2 sub-area images, and acquiring a portrait area image containing invasion pedestrians from the at least 2 sub-area images; and determining the position of the boundary-invading pedestrian according to the position of the area image of the portrait type in the space-based orbit image and the scene position corresponding to the space-based orbit image, thereby improving the accuracy of the detection of the boundary-invading pedestrian of the rail transit.
Description
Technical Field
The invention relates to a signal processing technology, in particular to a method, a device and a system for detecting an air-based pedestrian boundary violation of rail transit.
Background
The rail transit is a major artery of national economy, is the backbone of a comprehensive transportation network in China, and plays a very important role in a modern transportation system. Since railway tracks are mostly laid open air, there may be villages, fields or highways along the lines. Although isolation facilities such as fences and wire netting are usually arranged along the railway, the pedestrian boundary invasion phenomenon that pedestrians enter the range of the track often occurs, and in order to avoid collision accidents, the pedestrian boundary invasion needs to be rapidly detected so as to take braking measures in time or inform managers to stop the boundary invasion behaviors.
The existing pedestrian boundary intrusion detection method is that monitoring equipment such as a proximity detector and the like is usually buried along a railway, and when detecting that pedestrians and vehicles approach or enter a track area, an alarm is sent or alarm information is sent to a nearby management center.
However, with the development of a track system, a large number of detectors are arranged along a four-way eight-reach track, but the false alarm rate of the detectors is high, and the reliability of the existing pedestrian boundary intrusion detection method is not high.
Disclosure of Invention
The invention provides a method, a device and a system for detecting a space-based pedestrian boundary invasion of rail transit, which improve the accuracy and efficiency of detecting the pedestrian boundary invasion.
According to a first aspect of the invention, a method for detecting an air-based pedestrian boundary violation of rail transit is provided, which includes:
determining an anti-invasion boundary area image in a space-based orbit image according to the proportion of the pixel distance to the real distance in the space-based orbit image, wherein the space-based orbit image is an image shot by space-based image acquisition equipment vertically downwards along an orbit;
cutting the anti-invasion boundary area image into at least 2 sub-area images, and acquiring a portrait area image containing invasion pedestrians from the at least 2 sub-area images;
and determining the position of the boundary-invading pedestrian according to the position of the region image of the portrait type in the space-based orbit image and the scene position corresponding to the space-based orbit image.
Optionally, in a possible implementation manner of the first aspect, the determining, in the space-based orbit image, an anti-invasion boundary area image according to a ratio of a pixel distance to a true distance in the space-based orbit image includes:
judging whether the proportion of the pixel distance to the real distance in the space-based orbit image is larger than a proportion threshold value or not;
if yes, determining the empty base orbit image as an anti-invasion boundary area image;
if not, determining a selected frame of the anti-intrusion boundary area in the empty base orbit image by using a preset anti-intrusion boundary area detection model, and determining an image in the selected frame as the anti-intrusion boundary area image.
Optionally, in another possible implementation manner of the first aspect, before the determining, according to a ratio of a pixel distance to a true distance in the space-based orbit image, an anti-invasion boundary area image in the space-based orbit image, the method further includes:
acquiring a transverse pixel distance and a longitudinal pixel distance of the space-based track image;
acquiring shooting information corresponding to the space-based track image, wherein the shooting information comprises shooting height information and shooting visual angle information;
acquiring a transverse real distance and a longitudinal real distance corresponding to the space-based orbit image according to shooting information corresponding to the space-based orbit image;
determining a ratio of the lateral pixel distance to the lateral true distance as a first ratio;
determining a ratio of the longitudinal pixel distance to the longitudinal true distance as a second ratio;
and determining the minimum proportion of the first proportion and the second proportion as the proportion of the pixel distance to the real distance.
Optionally, in yet another possible implementation manner of the first aspect, before the determining, by the preset intrusion prevention area detection model, a selected frame of the intrusion prevention area in the base orbit image, the method further includes:
and performing fast-RCNN algorithm training on the anti-invasion boundary region detection model according to the real category of the historical invasion boundary region image, so that the loss value between the prediction category output by the anti-invasion boundary region detection model on the historical invasion boundary region image and the real category of the historical invasion boundary region image is lower than a preset fast-RCNN loss threshold value.
Optionally, in another possible implementation manner of the first aspect, the obtaining a human image type region map containing an encroaching pedestrian from the at least 2 sub-region maps includes:
acquiring HOG characteristic information of the direction gradient histogram of each sub-region graph;
determining HOG feature information of a portrait class in each HOG feature information by using a preset portrait detection model, wherein the portrait detection model is obtained by carrying out SVM (support vector machine) algorithm training on historical HOG feature information of the portrait class;
and determining the sub-region map corresponding to the human image HOG characteristic information as a human image region map containing boundary-invading pedestrians.
Optionally, in yet another possible implementation manner of the first aspect, before the determining, by using a preset portrait detection model, the portrait-like HOG feature information in each piece of HOG feature information, the method further includes:
if the proportion of the pixel distance to the real distance in the empty-base orbit image is larger than a proportion threshold value, carrying out SVM algorithm training on the portrait detection model according to the real class of the HOG feature information of a first historical portrait class, so that the loss value between the prediction class output by the portrait detection model on the HOG feature information of the first historical portrait class and the real class of the HOG feature information of the first historical portrait class is lower than a preset SVM loss threshold value, wherein the proportion of the pixel distance to the real distance in the historical empty-base orbit image corresponding to the HOG feature information of the first historical portrait class is larger than the proportion threshold value;
if the proportion of the pixel distance to the real distance in the empty-base orbit image is smaller than a proportion threshold value, carrying out SVM algorithm training on the portrait detection model according to the real class of second historical portrait HOG feature information, so that the loss value between the prediction class output by the portrait detection model on the second historical portrait HOG feature information and the real class of the second historical portrait HOG feature information is lower than a preset SVM loss threshold value, wherein the proportion of the pixel distance to the real distance in the historical empty-base orbit image corresponding to the second historical portrait HOG feature information is smaller than the proportion threshold value.
According to a second aspect of the present invention, there is provided a space-based pedestrian boundary intrusion detection device for rail transit, comprising:
the system comprises an anti-invasion boundary area image acquisition module, a boundary area image acquisition module and a boundary area image acquisition module, wherein the anti-invasion boundary area image acquisition module is used for determining an anti-invasion boundary area image in a space-based orbit image according to the proportion of the pixel distance to the real distance in the space-based orbit image, and the space-based orbit image is an image shot by space-based image acquisition equipment vertically downwards along an orbit;
the figure region image acquisition module is used for dividing the anti-invasion boundary region image into at least 2 sub-region images and acquiring a figure region image containing invasion pedestrians from the at least 2 sub-region images;
and the position determining module is used for determining the position of the boundary invading pedestrian according to the position of the region image of the portrait class in the space-based orbit image and the scene position corresponding to the space-based orbit image.
Optionally, in a possible implementation manner of the second aspect, the intrusion-prevention boundary area image obtaining module is specifically configured to determine whether a ratio of a pixel distance in the base orbit image to a true distance is greater than a ratio threshold; if yes, determining the empty base orbit image as an anti-invasion boundary area image; if not, determining a selected frame of the anti-intrusion boundary area in the empty base orbit image by using a preset anti-intrusion boundary area detection model, and determining an image in the selected frame as the anti-intrusion boundary area image.
Optionally, in another possible implementation manner of the second aspect, the anti-invasion boundary area image obtaining module is further configured to obtain a horizontal pixel distance and a vertical pixel distance of the empty base orbit image before determining the anti-invasion boundary area image in the empty base orbit image according to a ratio of the pixel distance in the empty base orbit image to a true distance; acquiring shooting information corresponding to the space-based track image, wherein the shooting information comprises shooting height information and shooting visual angle information; acquiring a transverse real distance and a longitudinal real distance corresponding to the space-based orbit image according to shooting information corresponding to the space-based orbit image; determining a ratio of the lateral pixel distance to the lateral true distance as a first ratio; determining a ratio of the longitudinal pixel distance to the longitudinal true distance as a second ratio; and determining the minimum proportion of the first proportion and the second proportion as the proportion of the pixel distance to the real distance.
Optionally, in yet another possible implementation manner of the second aspect, the anti-intrusion-boundary-region-image obtaining module is further configured to, before the determination of the selected frame of the anti-intrusion region in the space-based orbit image by using the preset anti-intrusion-boundary-region detection model, perform fast-RCNN algorithm training on the anti-intrusion-boundary-region detection model according to the true category of the historical anti-intrusion-boundary-region image, so that a loss value between a prediction category output by the anti-intrusion-boundary-region detection model on the historical anti-intrusion-boundary-region image and the true category of the historical anti-intrusion-boundary-region image is lower than a preset fast-RCNN loss threshold.
Optionally, in yet another possible implementation manner of the second aspect, the human image class region map obtaining module is specifically configured to obtain directional gradient histogram HOG feature information of each sub-region map; determining HOG feature information of a portrait class in each HOG feature information by using a preset portrait detection model, wherein the portrait detection model is obtained by carrying out SVM (support vector machine) algorithm training on historical HOG feature information of the portrait class; and determining the sub-region map corresponding to the human image HOG characteristic information as a human image region map containing boundary-invading pedestrians.
Optionally, in yet another possible implementation manner of the second aspect, the portrait type area map obtaining module is further configured to, before determining the portrait type HOG feature information in each HOG feature information by using a preset portrait detection model, perform SVM algorithm training on the portrait detection model according to a real category of first historical portrait type HOG feature information if a ratio of a pixel distance to the real distance in a space-based orbit image is greater than a proportional threshold, so that a loss value between a prediction category output by the portrait detection model on the first historical portrait type HOG feature information and the real category of the first historical portrait type HOG feature information is lower than a preset SVM loss threshold, where a ratio of a pixel distance to the real distance in a historical space-based orbit image corresponding to the first historical portrait type HOG feature information is greater than the proportional threshold; if the proportion of the pixel distance to the real distance in the empty-base orbit image is smaller than a proportion threshold value, carrying out SVM algorithm training on the portrait detection model according to the real class of second historical portrait HOG feature information, so that the loss value between the prediction class output by the portrait detection model on the second historical portrait HOG feature information and the real class of the second historical portrait HOG feature information is lower than a preset SVM loss threshold value, wherein the proportion of the pixel distance to the real distance in the historical empty-base orbit image corresponding to the second historical portrait HOG feature information is smaller than the proportion threshold value.
According to a third aspect of the invention, a space-based pedestrian boundary invasion detection system for rail transit is provided, which comprises a space-based pedestrian boundary invasion detection device for rail transit and at least one space-based image acquisition device;
the space-based image acquisition equipment is used for acquiring a space-based orbit image;
the air-based pedestrian boundary intrusion detection device for the rail transit is used for acquiring the air-based rail image from at least one air-based image acquisition device and executing the air-based pedestrian boundary intrusion detection method for the rail transit according to any one of claims 1 to 6.
According to a fourth aspect of the present invention, there is provided a space-based pedestrian boundary intrusion detection device for rail transit, comprising: memory, a processor and a computer program, the computer program being stored in the memory, the processor running the computer program to perform the method of the first aspect of the invention and its various possible designs.
According to a fifth aspect of the present invention, there is provided a readable storage medium having stored therein a computer program for implementing the method of the first aspect of the present invention and its various possible designs when executed.
According to the method, the device and the system for detecting the space-based pedestrian boundary invasion, provided by the invention, a boundary invasion prevention area image is determined in a space-based track image according to the proportion of the pixel distance to the real distance in the space-based track image, and the space-based track image is an image vertically shot downwards along a track by space-based image acquisition equipment; cutting the anti-invasion boundary area image into at least 2 sub-area images, and acquiring a portrait area image containing invasion pedestrians from the at least 2 sub-area images; and determining the position of the boundary-invading pedestrian according to the position of the area image of the portrait type in the space-based orbit image and the scene position corresponding to the space-based orbit image, thereby improving the accuracy of the detection of the boundary-invading pedestrian of the rail transit.
Drawings
Fig. 1 is a schematic view of an application scenario of a space-based pedestrian boundary intrusion detection system for rail transit according to an embodiment of the present invention;
fig. 2 is a schematic flowchart of a method for detecting an air-based pedestrian boundary violation in rail transit according to an embodiment of the present invention;
fig. 3 is a schematic flowchart of another method for detecting an air-based pedestrian boundary violation in rail transit according to an embodiment of the present invention;
fig. 4 is a schematic flowchart of a method for detecting an air-based pedestrian boundary violation in rail transit according to another embodiment of the present invention;
fig. 5 is a schematic structural diagram of an air-based pedestrian boundary intrusion detection device for rail transit according to an embodiment of the present invention;
fig. 6 is a schematic hardware structure diagram of an air-based pedestrian boundary intrusion detection device for rail transit according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The terms "first," "second," "third," "fourth," and the like in the description and in the claims, as well as in the drawings, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein.
It should be understood that, in the various embodiments of the present application, the size of the serial number of each process does not mean the execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.
It should be understood that, in this application, "comprises" and "comprising," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
It should be understood that in this application, "plurality" means two or more. "and/or" is merely an association describing an associated object, meaning that three relationships may exist, for example, and/or B, may mean: a exists alone, A and B exist simultaneously, and B exists alone. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship. "comprising A, B and C" means A, B, C includes all three,
it should be understood that in the present application, "B corresponding to a", "a corresponds to B", or "B corresponds to a" means that B is associated with a, from which B can be determined. Determining B from a does not mean determining B from a alone, but may be determined from a and/or other information. And the matching of A and B means that the similarity of A and B is greater than or equal to a preset threshold value.
It should be understood that in the present application, the fast-RCNN algorithm is an object detection algorithm based on a convolutional neural network. The candidate regions (regions) are determined by the Fast-RCNN through a neural network, the Fast-RCNN uses a shared convolution layer mode on the basis of the Fast-RCNN, and the convolved feature map is used for generating the candidate regions. A candidate Region extraction network (RPNs) is realized by adding two convolutional layers on the basis of Fast-RCNN, wherein one convolutional layer is used for coding the position of each feature map into a vector, and the other convolutional layer outputs a objective score (objective score) and regression boundaries (regression boundaries for k candidate regions) for each position.
It should be understood that, in the present application, the ZF-net algorithm is another algorithm based on the convolutional neural network, and is characterized by a fine tuning algorithm based on Alex-net, using the ReLU activation function, the cross entropy cost function, and using a smaller filter (filter) to retain more original pixel information.
It should be understood that, in the present application, svm (support Vector machine) refers to a support Vector machine, which is a common discriminant method. In the field of machine learning, a supervised learning model is typically used for pattern recognition, classification, and regression analysis.
It should be understood that, in the present application, the space-based orbit image is a space-based image in which an orbit image is taken. The space-based image refers to an image with a non-fixed background, and is generally an image shot by a moving camera. In the continuously shot space-based images, both the background and the shot object are changed.
It should be understood that, in the present application, the HOG feature information refers to Histogram of Oriented Gradient (HOG) feature information, which is a feature descriptor used for object detection in computer vision and image processing. The HOG feature information constitutes features by calculating and counting a gradient direction histogram of a local region of an image. In one image, the appearance and shape of the local object can be well described by the directional density distribution of the gradient or edge. The essence is as follows: statistics of the gradient, while the gradient is mainly present at the edge. The realization method mainly comprises the following steps: the image is first divided into small connected regions, which are called cell units. And then acquiring the gradient or edge direction histogram of each pixel point in the cell unit. Finally, combining these histograms can form a feature descriptor, i.e. the HOG feature information of the present application. The performance of the algorithm can be improved by performing contrast normalization on the local histograms in a larger range (called interval) of the image, and the method is as follows: the density of each histogram in this bin is calculated and then each cell unit in the bin is normalized based on this density. By this normalization, better effects on illumination variations and shadows can be obtained.
As used herein, "if" may be interpreted as "at … …" or "when … …" or "in response to a determination" or "in response to a detection", depending on the context.
The technical solution of the present invention will be described in detail below with specific examples. The following several specific embodiments may be combined with each other, and details of the same or similar concepts or processes may not be repeated in some embodiments.
Fig. 1 is a schematic view of an application scenario of a space-based pedestrian boundary intrusion detection system for rail transit according to an embodiment of the present invention. In the application scenario shown in fig. 1, the air-based pedestrian boundary intrusion detection system for rail transit mainly includes an air-based pedestrian boundary intrusion detection device for rail transit and at least one air-based image acquisition device. The air-based pedestrian boundary-crossing detection device of the rail transit can be understood as a server 1 shown in fig. 1, and the air-based image acquisition equipment can be understood as an unmanned aerial vehicle 2 shown in fig. 1. Unmanned aerial vehicle 2 flies directly over along orbital, and unmanned aerial vehicle's camera 21 sets up under its organism, can shoot the sky base track image perpendicularly downwards from this. Therefore, the image center of the aerial orbit image should move along the central axis of the orbit, but since the flying height of the unmanned aerial vehicle relative to the ground may be changed, some aerial orbit images only contain the anti-invasion boundary area of the orbit, and some aerial orbit images contain the anti-invasion boundary area of the orbit and the scene images at two sides of the orbit. According to the air-based pedestrian boundary intrusion detection system for the rail transit, provided by the embodiment of the invention, the air-based image acquisition equipment is used for acquiring an air-based rail image, and the air-based pedestrian boundary intrusion detection device for the rail transit is used for acquiring the air-based rail image from at least one air-based image acquisition equipment and executing the air-based pedestrian boundary intrusion detection method for the rail transit in the following method embodiments, so that the detection accuracy and efficiency of the pedestrian boundary intrusion are improved.
Referring to fig. 2, which is a schematic flowchart of a method for detecting an air-based pedestrian boundary violation in rail transit according to an embodiment of the present invention, an execution main body of the method shown in fig. 2 may be a software and/or hardware device. The method shown in fig. 2 mainly includes steps S101 to S103, which are as follows:
s101, determining an anti-invasion boundary area image in the empty base track image according to the proportion of the pixel distance to the real distance in the empty base track image, wherein the empty base track image is an image shot by empty base image acquisition equipment vertically downwards along a track.
Specifically, the intrusion prevention area may be an area surrounded by barriers, wire netting, and other isolation facilities on both sides of the rail, or may be a preset area that is radially spaced from the rail by a certain distance. The proportion of the pixel distance to the real distance in the space-based orbit image represents whether the space-based orbit image is shot by the space-based image acquisition equipment close to the ground or far away from the ground. When the proportion of the pixel distance to the real distance is larger, the same pedestrian is a large-size pedestrian target in the space-based track image, for example, the portrait occupies half the area of the picture; and when the proportion of the pixel distance to the real distance is small, the pedestrian in the space-based track image is a small-size pedestrian target, for example, the portrait occupies 1/5 area of the picture. Because under the condition of shooting far away from the ground, the empty foundation track image is easy to introduce more non-track area images, such as objects on the two sides of the track area, such as rice fields, stones and the like, which may have adverse effects on detection, and the detection accuracy is reduced. In the embodiment, under the condition that the proportion of the pixel distance to the real distance in the space-based orbit image is different, the corresponding anti-invasion boundary area image is adopted, so that the accuracy of the anti-invasion boundary area image is improved.
It can be understood that a preset image processing rule corresponding to the proportion is obtained from the proportion of the pixel distance to the real distance in the empty base orbit image, so as to determine an anti-invasion boundary area image in the empty base orbit image. For example, the ratio of each pixel distance to the real distance corresponds to a screenshot size, and image capture of bilateral symmetric regions is performed on the basis of the central axis of the picture in the space-based track image according to the corresponding screenshot size to obtain a corresponding anti-intrusion boundary region image. For example, when the ratio of the pixel distance to the real distance is greater than a preset threshold, the empty base orbit image is directly determined as an anti-invasion boundary area image; and when the ratio of the pixel distance to the real distance is smaller than or equal to a preset threshold value, carrying out image identification on the anti-invasion boundary area in the empty base track image, and determining the identified area as the anti-invasion boundary area image. In this embodiment, the proportion of the pixel distance to the real distance in the empty base orbit image is used as a basis to determine the intrusion-prevention boundary area image from the empty base orbit image, and various implementation manners are not limited herein.
S102, the anti-invasion boundary area image is divided into at least 2 sub-area images, and a portrait area image containing the invasion pedestrians is obtained from the at least 2 sub-area images.
Specifically, since the pedestrian who has the boundary encroaching behavior may be located at any position of the boundary encroaching region, the image of the boundary encroaching region may be obtained by dividing the image of the boundary encroaching region into at least 2 sub-region maps and detecting whether the sub-regions are the image of the person one by one. For example, if it is detected that all the adjacent 2 sub-region maps are the portrait region map, it indicates that the photographed pedestrian is in the adjacent 2 sub-region maps.
S103, determining the position of the boundary-invading pedestrian according to the position of the region diagram of the portrait in the space-based orbit image and the scene position corresponding to the space-based orbit image.
Specifically, each time the space-based image acquisition device acquires one space-based orbit image, the current acquisition position can be acquired, and the acquisition position is used as the scene position corresponding to the space-based orbit image. The position range of the pedestrian invading the boundary can be preliminarily determined according to the scene position corresponding to the empty base orbit image, for example, the position range is within the range of the track section along the route with longitude XX latitude YY as the symmetric center and radius of 200 meters. And then obtaining the relative position of the boundary-invading pedestrian in the range of the road section along the track according to the position of the region image of the portrait in the image of the empty base track, thereby accurately obtaining the position of the boundary-invading pedestrian.
The embodiment of the invention provides a method for detecting an intrusion boundary of an empty-base pedestrian in rail transit, which comprises the steps of determining an anti-intrusion boundary area image in an empty-base rail image according to the proportion of the pixel distance to the real distance in the empty-base rail image, wherein the empty-base rail image is an image vertically shot downwards along a rail by an empty-base image acquisition device; cutting the anti-invasion boundary area image into at least 2 sub-area images, and acquiring a portrait area image containing invasion pedestrians from the at least 2 sub-area images; and determining the position of the boundary-invading pedestrian according to the position of the area image of the portrait type in the space-based orbit image and the scene position corresponding to the space-based orbit image, thereby improving the accuracy of the detection of the boundary-invading pedestrian of the rail transit.
Referring to fig. 3, which is a schematic flow chart of another method for detecting an intrusion boundary of a space-based pedestrian in rail transit according to an embodiment of the present invention, on the basis of the above embodiment, step S101 (determining an intrusion-prevention boundary area image in a space-based orbit image according to a ratio of a pixel distance to a true distance in the space-based orbit image) may be replaced with steps S201 to S204 shown in fig. 3, specifically as follows:
s201, judging whether the proportion of the pixel distance in the space-based orbit image to the real distance is larger than a proportion threshold value.
If so, the process proceeds to S202, and if not, the process proceeds to S203.
Specifically, the proportion of the pixel distance to the real distance in the space-based orbit image is greater than a proportion threshold value, which indicates that the space-based orbit image is a large-size image and is shot near the ground, and the shot scene is an anti-invasion boundary area image. And the proportion of the pixel distance to the real distance in the empty base orbit image is smaller than or equal to a proportion threshold value, which indicates that the empty base orbit image is a small-size image and needs to be shot far away from the ground, the shot scene also shoots scenes on two sides of the orbit area besides the anti-invasion boundary area, image clipping is needed, and the anti-invasion boundary area image in the empty base orbit image is extracted.
S202, determining the empty base orbit image as an anti-invasion boundary area image.
S203, determining a selection frame of the anti-invasion boundary area in the empty base track image by using a preset anti-invasion boundary area detection model.
Specifically, the preset intrusion prevention boundary area detection model may be a fast-RCNN model. The fast-RCNN model can be used for detecting objects in the images, for example, if the objects to be detected in the positive samples of the training set are all anti-invasion boundary areas of the orbit, and the negative samples are peripheral environment images of the non-anti-invasion boundary areas, the trained fast-RCNN model can be used for framing the anti-invasion boundary areas in the air-based orbit image.
Before step S203 (using a preset anti-intrusion boundary area detection model to determine a selection frame of the anti-intrusion boundary area in the empty base orbit image), an anti-intrusion boundary area detection model training process may be further included. Specifically, the fast-RCNN algorithm training may be performed on the anti-invasion boundary area detection model according to the real category of the historical invasion boundary area image, so that the loss value between the prediction category output by the anti-invasion boundary area detection model on the historical invasion boundary area image and the real category of the historical invasion boundary area image is lower than a preset fast-RCNN loss threshold.
In a specific practical mode, before a space-based track image is detected, an anti-invasion boundary area training sample in a training image library is used for training an anti-invasion boundary area detection model. The training image library may be derived from pre-collected rail transit drone patrol videos. Firstly, manually marking training pictures, and marking which pictures are anti-invasion boundary area pictures and which pictures are other pictures to obtain an anti-invasion boundary area training sample. Alternatively, in selecting a positive sample, a maximum rectangular frame image containing only the intrusion prevention area, and not the background area (farmland, road, etc.) may be selected. When an included angle exists between the track direction in the picture and the picture transverse axis direction (namely the track is in a curved shape), the picture can be cut off according to the track shape, and a plurality of straight-line-segment tracks are cut out and respectively used as independent positive samples, so that the area of a background region in the rectangular frame is reduced as much as possible. The negative sample can be selected by randomly selecting the image of the background area (farmland, road, etc.). And the anti-invasion boundary region detection model adopts a fast-RCNN model, and after the training set samples are labeled, each training picture is sequentially input into the fast-RCNN model to carry out model learning and training for many times. The Fast-RCNN model mainly comprises two convolutional neural networks, namely a candidate region extraction network (RPN) and a Fast-RCNN detection network. The method for alternately training the two networks is adopted during training of the Faster-RCNN model, the RPN is trained, the Fast-RCNN detection network is trained by using the obtained candidate region, then the shared depth convolution layer of the RPN and the Fast-RCNN detection network is fixed, the full connection layer of the RPN is optimized, and finally the full connection layer of the detection network is optimized. The RPN and Fast-RCNN detection networks both need pre-training of an initialization network, and a specific layer is added after the network output is initialized. The pre-training initialization network of the present embodiment may be a ZFnet network. The ZFnet network comprises 5 convolutional layers, 3 pooling layers and 3 fully-connected layers, and the output of the 5 th convolutional layer of ZFnet is called a feature map. When a Fast-RCNN model is trained by using an anti-invasion boundary area sample library, the RPN and the Fast-RCNN detection network can be optimized by directly using ZFNet initialization parameters obtained by pre-training.
In the detection process of the fast-RCNN model after training, a series of convolution operations are performed on the whole space-based orbit image to obtain a feature map of an input picture, and then the feature map is input into an RPN network to generate candidate regions (regions), for example, about 300 candidate regions are generated. And then, taking the characteristic diagram and the output of the RPN as the input of the pooling layer to obtain fixed-length output. And finally, classifying the full connection layer and the softmax layer to obtain a detection result of the selected frame for indicating the anti-invasion area.
S204, determining the image in the selected frame as the anti-invasion boundary area image.
Specifically, the image in the selected frame is the image range of the anti-intrusion boundary area identified by the fast-RCNN model in the space-based orbit image, and therefore the image in the selected frame is intercepted, namely the anti-intrusion boundary area image.
On the basis of the above embodiment, before step S101 (determining an anti-intrusion boundary area image in the empty base orbit image according to the ratio of the pixel distance to the true distance in the empty base orbit image), a process of acquiring the ratio of the pixel distance to the true distance may be further included, specifically: firstly, the horizontal pixel distance and the longitudinal pixel distance of the space-based track image are obtained. The lateral pixel distance is understood to be the number of pixels in the lateral direction of the image of the empty base track, or the distance on the image. Similarly, the longitudinal pixel distance may be understood as the number of pixels in the longitudinal direction of the image of the empty base track or the distance on the image, but is consistent with the unit of the transverse pixel distance (i.e., the transverse pixel distance and the longitudinal pixel distance are both the number of pixels or both the distance). Acquiring shooting information corresponding to the space-based track image while acquiring a transverse pixel distance and a longitudinal pixel distance, or before or after acquiring the transverse pixel distance and the longitudinal pixel distance, wherein the shooting information comprises shooting height information and shooting visual angle information; and acquiring a transverse real distance and a longitudinal real distance corresponding to the space-based orbit image according to the shooting information corresponding to the space-based orbit image. The shooting height information can be understood as the distance between a camera of the space-based image acquisition equipment and the ground of a track area right below the camera, and can be obtained from device parameters of the space-based image acquisition equipment. The shooting angle of view information may be understood as a lens angle of view, such as a horizontal angle of view and a vertical angle of view, of the image captured by the space-based image capturing apparatus. After the distance is obtained, determining the proportion of the transverse pixel distance to the transverse true distance as a first proportion; and determining a ratio of the longitudinal pixel distance to the longitudinal true distance as a second ratio. The minimum ratio of the first ratio and the second ratio may be determined as a ratio of the pixel distance to a real distance.
In one implementation, the process of obtaining the ratio of the pixel distance to the real distance may be: and acquiring a visible light video of the space-based image acquisition equipment for routing inspection along the railway, and extracting a space-based track image from the video. The space-based image acquisition device is, for example, an unmanned aerial vehicle, and in the case of vertical downward view, the flying height H of the unmanned aerial vehicle is acquired, the size of the shot image is w × H pixels, and the angle of view of the camera is α × β degrees (horizontal × vertical). Thus, the lateral true distance can be found as: l ishor2H · tan (α/2), the true longitudinal distance is: l isver2H · tan (β/2). Transverse in the imageAnd the actual geographic distance represented by the unit pixel in the longitudinal direction is respectively the inverse of the first ratio: phor＝LhorW and the inverse of the second ratio: pver＝LverH is used as the reference value. Suppose PthFor a predetermined ratio threshold, for example, P can be setth0.015, if min (P)hor,Pver)＞PthIndicating that the size of the pedestrian object in the image is small; otherwise, it indicates that the size of the pedestrian object in the image is large.
Referring to fig. 4, which is a schematic flow chart of another method for detecting an intrusion of an air-based pedestrian in rail transit according to an embodiment of the present invention, on the basis of the above embodiment, the process of acquiring a portrait type area map including an intrusion pedestrian from the at least 2 sub-area maps in step S102 may be replaced with steps S301 to S303 shown in fig. 4, specifically as follows:
s301, acquiring HOG characteristic information of the direction gradient histograms of the sub-region maps.
Specifically, in the acquisition process of the HOG feature information, it may be set that a gray value at a pixel point (x, y) of the sub-region map is I, a gradient amplitude is G, and a gradient direction is θ, and a one-dimensional central gradient operator of [ -1,0,1] is adopted to calculate gradients in the horizontal direction and the vertical direction as follows:
Gx(x,y)＝I(x+1,y)-I(x-1,y)，
Gy(x,y)＝I(x,y+1)-I(x,y-1)。
then, the gradient strength of the pixel point (x, y) is calculated:
and the gradient direction of the pixel point (x, y):
and for the RGB color image, calculating the gradient of each color channel, and selecting the gradient corresponding to the color channel with the maximum gradient amplitude as the gradient of the pixel point. The sub-region map is then divided into evenly distributed cells (cells), for example of 8 x 8 pixel size, where each adjacent 2 x 2 cell is grouped into one small region block. And obtaining pixels by taking the gradient strength as a weight in the small area block. The positive and negative of the gradient direction are not considered, namely the direction is converted into 0-180 degrees, and 9 direction columns are taken as a histogram. The histogram of each small region block is normalized, thereby enabling better invariance to illumination, shadows, edge contrast, and the like. Because each small region block has 4 9-dimensional histograms, a 36-dimensional feature vector of the small region block is obtained after normalization. Assuming v is a normalized feature vector, the common normalization method steps are step 1 to step 2:
And 2, reducing the element to 0.2 if the element is larger than 0.2 according to the result of the step 1, then executing the step 1 again until all the elements are smaller than or equal to 0.2, and ending the normalization.
After the normalization operation, the feature vectors of all small region blocks are connected in series to form the HOG features of the subregion map.
S302, determining HOG feature information of a portrait class in each HOG feature information by using a preset portrait detection model, wherein the portrait detection model is obtained by using historical HOG feature information to carry out SVM algorithm training.
Specifically, the portrait detection model may be an SVM model, and the HOG features are input into the SVM model for detection, and the SVM model is a linear SVM model. The specific mode for determining the image type HOG feature information in each HOG feature information can be a mode of sliding a window, sliding a data acquisition window in the air-based orbit image, sequentially taking each sub-region image as a window image, extracting HOG features in the window and inputting the HOG features into an SVM (support vector machine) model for classification, so that the detection and the positioning of the boundary-invaded pedestrians in the air-based orbit image are realized.
Before step S302 (using a preset portrait detection model to determine the portrait type HOG feature information in each piece of HOG feature information), training of a portrait detection model is further included in the following two implementation manners:
in one implementation, if the ratio of the pixel distance to the real distance in the space-based orbit image is greater than a ratio threshold, the space-based orbit image is represented as a large-size image, and therefore an SVM model trained by the same large-size image is used for detection. Specifically, SVM algorithm training is carried out on the portrait detection model according to the real category of the HOG feature information of the first historical portrait type, so that the loss value between the prediction category output by the portrait detection model on the HOG feature information of the first historical portrait type and the real category of the HOG feature information of the first historical portrait type is lower than a preset SVM loss threshold value, wherein the proportion of the pixel distance to the real distance in the historical space-based orbit image corresponding to the HOG feature information of the first historical portrait type is larger than a proportion threshold value.
In another implementation mode, if the proportion of the pixel distance to the real distance in the space-based orbit image is smaller than a proportion threshold value, the space-based orbit image is represented as a small-size image, and therefore an SVM model obtained by training the small-size image is used for detection. Specifically, SVM algorithm training is carried out on the portrait detection model according to the real category of the HOG feature information of the second historical portrait type, so that the loss value between the prediction category output by the portrait detection model on the HOG feature information of the second historical portrait type and the real category of the HOG feature information of the second historical portrait type is lower than a preset SVM loss threshold value, wherein the proportion of the pixel distance to the real distance in the historical space-based orbit image corresponding to the HOG feature information of the second historical portrait type is smaller than a proportional threshold value.
Since the feature quantities that the large-size and small-size figures can provide have a large difference, if the SVM model obtained by the large-size figure training set is used for detecting the small-size figures in the space-based orbit image, the missing rate is high, in the above embodiment, the size grade (for example, large size or small size) of the figures in the image is determined according to the ratio of the pixel distance in the space-based orbit image to the real distance, and then the SVM model obtained by the image training of the size grade is used for detecting, so that the accuracy of detecting the invaded pedestrians is improved.
Optionally, a cross validation method may be used to select an optimal SVM parameter in the training of the SVM model, so as to improve the classification accuracy.
And S303, determining the sub-region map corresponding to the human image HOG characteristic information as a human image region map containing boundary-invading pedestrians.
Specifically, the SVM model in this embodiment recognizes only the feature information of the image type HOG, and does not distinguish between 1 person and 2 persons. Therefore, if the sub-region map corresponding to the human image type HOG feature information is a sub-region map of multiple persons, the sub-region map corresponding to the human image type HOG feature information may be a neighboring sub-region map or may be a non-neighboring sub-region map. Correspondingly, the positions of all the sub-region maps corresponding to the human image type HOG characteristic information are used as the positions of the boundary-invading pedestrians.
Referring to fig. 5, a schematic structural diagram of a device for detecting an air-based pedestrian boundary violation in rail transit provided by an embodiment of the present invention mainly includes:
the anti-invasion boundary area image acquisition module 51 is configured to determine an anti-invasion boundary area image in the empty base orbit image according to a ratio of a pixel distance in the empty base orbit image to a real distance, where the empty base orbit image is an image that is shot by an empty base image acquisition device vertically and downwards along an orbit;
the portrait type area map acquisition module 52 is configured to divide the intrusion prevention area image into at least 2 sub-area maps, and acquire a portrait type area map including an intrusion pedestrian from the at least 2 sub-area maps;
and the position determining module 53 is configured to determine the position of the boundary encroaching pedestrian according to the position of the region map of the portrait type in the space-based orbit image and the scene position corresponding to the space-based orbit image.
The air-based pedestrian boundary-crossing detection device for rail transit shown in the embodiment of fig. 5 can be correspondingly used for executing the steps in the method embodiment shown in fig. 2, and the implementation principle and the technical effect are similar, and are not described again here.
On the basis of the above embodiment, the anti-intrusion boundary area image obtaining module 51 is specifically configured to determine whether a ratio of a pixel distance in the empty base orbit image to a real distance is greater than a ratio threshold; if yes, determining the empty base orbit image as an anti-invasion boundary area image; if not, determining a selected frame of the anti-intrusion boundary area in the empty base orbit image by using a preset anti-intrusion boundary area detection model, and determining an image in the selected frame as the anti-intrusion boundary area image.
On the basis of the foregoing embodiment, the anti-intrusion boundary area image obtaining module 51 is further configured to obtain a horizontal pixel distance and a vertical pixel distance of the empty base track image before determining the anti-intrusion boundary area image in the empty base track image according to a ratio of the pixel distance to a real distance in the empty base track image; acquiring shooting information corresponding to the space-based track image, wherein the shooting information comprises shooting height information and shooting visual angle information; acquiring a transverse real distance and a longitudinal real distance corresponding to the space-based orbit image according to shooting information corresponding to the space-based orbit image; determining a ratio of the lateral pixel distance to the lateral true distance as a first ratio; determining a ratio of the longitudinal pixel distance to the longitudinal true distance as a second ratio; and determining the minimum proportion of the first proportion and the second proportion as the proportion of the pixel distance to the real distance.
On the basis of the foregoing embodiment, the anti-intrusion boundary area image obtaining module 51 is further configured to, before the preset anti-intrusion boundary area detection model is used to determine the selected frame of the anti-intrusion boundary area in the space-based orbit image, perform fast-RCNN algorithm training on the anti-intrusion boundary area detection model according to the real category of the historical anti-intrusion boundary area image, so that a loss value between a prediction category output by the anti-intrusion boundary area detection model on the historical anti-intrusion boundary area image and the real category of the historical anti-intrusion boundary area image is lower than a preset fast-RCNN loss threshold.
On the basis of the foregoing embodiment, the human image type region map obtaining module 52 is specifically configured to obtain directional gradient histogram HOG feature information of each sub-region map; determining HOG feature information of a portrait class in each HOG feature information by using a preset portrait detection model, wherein the portrait detection model is obtained by carrying out SVM (support vector machine) algorithm training on historical HOG feature information of the portrait class; and determining the sub-region map corresponding to the human image HOG characteristic information as a human image region map containing boundary-invading pedestrians.
On the basis of the foregoing embodiment, the human image type region map obtaining module 52 is further configured to, before determining the human image type HOG feature information in each HOG feature information by using a preset human image detection model, if a ratio of a pixel distance to a real distance in a space-based orbit image is greater than a ratio threshold, perform SVM algorithm training on the human image detection model by using a real category of first historical human image type HOG feature information, so that a loss value between a prediction category output by the human image detection model for the first historical human image type HOG feature information and the real category of the first historical human image type HOG feature information is lower than a preset loss threshold, where a ratio of a pixel distance to a real distance in a historical space-based orbit image corresponding to the first historical human image type HOG feature information is greater than a ratio threshold; if the proportion of the pixel distance to the real distance in the empty-base orbit image is smaller than a proportion threshold value, carrying out SVM algorithm training on the portrait detection model according to the real class of second historical portrait HOG feature information, so that the loss value between the prediction class output by the portrait detection model on the second historical portrait HOG feature information and the real class of the second historical portrait HOG feature information is lower than a preset SVM loss threshold value, wherein the proportion of the pixel distance to the real distance in the historical empty-base orbit image corresponding to the second historical portrait HOG feature information is smaller than the proportion threshold value.
Referring to fig. 6, which is a schematic diagram of a hardware structure of a space-based pedestrian boundary intrusion detection device for rail transit provided in an embodiment of the present invention, the device includes: a processor 61, memory 62 and computer programs; wherein
A memory 62 for storing the computer program, which may also be a flash memory (flash). The computer program is, for example, an application program, a functional module, or the like that implements the above method.
A processor 61 for executing the computer program stored in the memory to implement the steps performed by the apparatus in the above method. Reference may be made in particular to the description relating to the preceding method embodiment.
Alternatively, the memory 62 may be separate or integrated with the processor 61.
When the memory 62 is a device independent of the processor 61, the apparatus may further include:
a bus 63 for connecting the memory 62 and the processor 61.
The present invention also provides a readable storage medium, in which a computer program is stored, which when executed is used to implement the methods provided by the various embodiments described above.
The readable storage medium may be a computer storage medium or a communication medium. Communication media includes any medium that facilitates transfer of a computer program from one place to another. Computer storage media may be any available media that can be accessed by a general purpose or special purpose computer. For example, a readable storage medium is coupled to the processor such that the processor can read information from, and write information to, the readable storage medium. Of course, the readable storage medium may also be an integral part of the processor. The processor and the readable storage medium may reside in an Application Specific Integrated Circuits (ASIC). Additionally, the ASIC may reside in user equipment. Of course, the processor and the readable storage medium may also reside as discrete components in a communication device.
The present invention also provides a program product comprising execution instructions stored in a readable storage medium. The at least one processor of the device may read the execution instructions from the readable storage medium, and the execution of the execution instructions by the at least one processor causes the device to implement the methods provided by the various embodiments described above.
In the above embodiments of the air-based pedestrian boundary intrusion detection device for rail transit, it should be understood that the Processor may be a Central Processing Unit (CPU), other general-purpose processors, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of a method disclosed in connection with the present application may be embodied directly in a hardware processor, or in a combination of the hardware and software modules in the processor.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.
Claims (6)
1. A method for detecting an air-based pedestrian boundary violation of rail transit is characterized by comprising the following steps:
determining an anti-invasion boundary area image in a space-based orbit image according to the proportion of the pixel distance to the real distance in the space-based orbit image, wherein the space-based orbit image is an image shot by space-based image acquisition equipment vertically downwards along an orbit;
cutting the anti-invasion boundary area image into at least 2 sub-area images, and acquiring a portrait area image containing invasion pedestrians from the at least 2 sub-area images;
determining the position of the boundary-invading pedestrian according to the position of the area image of the portrait type in the space-based orbit image and the scene position corresponding to the space-based orbit image, wherein the scene position is the acquisition position of the space-based orbit image;
determining an anti-invasion boundary area image in the space-based orbit image according to the proportion of the pixel distance to the real distance in the space-based orbit image, wherein the method comprises the following steps:
judging whether the proportion of the pixel distance to the real distance in the space-based orbit image is larger than a proportion threshold value or not;
if yes, determining the empty base orbit image as an anti-invasion boundary area image;
if not, determining a selected frame of the anti-intrusion boundary area in the empty base track image by using a preset anti-intrusion boundary area detection model, and determining an image in the selected frame as the anti-intrusion boundary area image;
before determining an anti-invasion boundary area image in the space-based orbit image according to the proportion of the pixel distance to the true distance in the space-based orbit image, the method further comprises the following steps:
acquiring a transverse pixel distance and a longitudinal pixel distance of the space-based track image;
acquiring shooting information corresponding to the space-based track image, wherein the shooting information comprises shooting height information and shooting visual angle information;
acquiring a transverse real distance and a longitudinal real distance corresponding to the space-based orbit image according to shooting information corresponding to the space-based orbit image;
determining a ratio of the lateral pixel distance to the lateral true distance as a first ratio;
determining a ratio of the longitudinal pixel distance to the longitudinal true distance as a second ratio;
and determining the minimum proportion of the first proportion and the second proportion as the proportion of the pixel distance to the real distance.
2. The method according to claim 1, wherein before determining the selected frame of the anti-invasion area in the image of the empty base orbit by using the preset anti-invasion area detection model, the method further comprises:
and performing fast-RCNN algorithm training on the anti-invasion boundary region detection model according to the real category of the historical invasion boundary region image, so that the loss value between the prediction category output by the anti-invasion boundary region detection model on the historical invasion boundary region image and the real category of the historical invasion boundary region image is lower than a preset fast-RCNN loss threshold value.
3. The method according to claim 1, wherein the obtaining the portrait area map containing the encroaching pedestrian from the at least 2 sub-area maps comprises:
acquiring HOG characteristic information of the direction gradient histogram of each sub-region graph;
determining HOG feature information of a portrait class in each HOG feature information by using a preset portrait detection model, wherein the portrait detection model is obtained by carrying out SVM (support vector machine) algorithm training on historical HOG feature information of the portrait class;
and determining the sub-region map corresponding to the human image HOG characteristic information as a human image region map containing boundary-invading pedestrians.
4. The method according to claim 3, wherein before determining the HOG feature information of human image class in each HOG feature information by using the preset human image detection model, the method further comprises:
if the proportion of the pixel distance to the real distance in the empty-base orbit image is larger than a proportion threshold value, carrying out SVM algorithm training on the portrait detection model according to the real class of the HOG feature information of a first historical portrait class, so that the loss value between the prediction class output by the portrait detection model on the HOG feature information of the first historical portrait class and the real class of the HOG feature information of the first historical portrait class is lower than a preset SVM loss threshold value, wherein the proportion of the pixel distance to the real distance in the historical empty-base orbit image corresponding to the HOG feature information of the first historical portrait class is larger than the proportion threshold value;
if the proportion of the pixel distance to the real distance in the empty-base orbit image is smaller than a proportion threshold value, carrying out SVM algorithm training on the portrait detection model according to the real class of second historical portrait HOG feature information, so that the loss value between the prediction class output by the portrait detection model on the second historical portrait HOG feature information and the real class of the second historical portrait HOG feature information is lower than a preset SVM loss threshold value, wherein the proportion of the pixel distance to the real distance in the historical empty-base orbit image corresponding to the second historical portrait HOG feature information is smaller than the proportion threshold value.
5. The utility model provides a space-based pedestrian invades border detection device of track traffic which characterized in that includes:
the system comprises an anti-invasion boundary area image acquisition module, a boundary area image acquisition module and a boundary area image acquisition module, wherein the anti-invasion boundary area image acquisition module is used for determining an anti-invasion boundary area image in a space-based orbit image according to the proportion of the pixel distance to the real distance in the space-based orbit image, and the space-based orbit image is an image shot by space-based image acquisition equipment vertically downwards along an orbit;
the figure region image acquisition module is used for dividing the anti-invasion boundary region image into at least 2 sub-region images and acquiring a figure region image containing invasion pedestrians from the at least 2 sub-region images;
the position determining module is used for determining the position of the boundary-invading pedestrian according to the position of the area image of the portrait class in the space-based orbit image and the scene position corresponding to the space-based orbit image, wherein the scene position is the acquisition position of the space-based orbit image;
the anti-invasion boundary area image acquisition module is specifically used for judging whether the proportion of the pixel distance to the real distance in the empty base orbit image is greater than a proportion threshold value; if yes, determining the empty base orbit image as an anti-invasion boundary area image; if not, determining a selected frame of the anti-intrusion boundary area in the empty base track image by using a preset anti-intrusion boundary area detection model, and determining an image in the selected frame as the anti-intrusion boundary area image;
the anti-invasion boundary area image acquisition module is further configured to acquire a transverse pixel distance and a longitudinal pixel distance of the empty base orbit image before determining the anti-invasion boundary area image in the empty base orbit image according to the proportion of the pixel distance to the real distance in the empty base orbit image; acquiring shooting information corresponding to the space-based track image, wherein the shooting information comprises shooting height information and shooting visual angle information; acquiring a transverse real distance and a longitudinal real distance corresponding to the space-based orbit image according to shooting information corresponding to the space-based orbit image; determining a ratio of the lateral pixel distance to the lateral true distance as a first ratio; determining a ratio of the longitudinal pixel distance to the longitudinal true distance as a second ratio; and determining the minimum proportion of the first proportion and the second proportion as the proportion of the pixel distance to the real distance.
6. The air-based pedestrian boundary invasion detection system for rail transit is characterized by comprising an air-based pedestrian boundary invasion detection device for rail transit and at least one air-based image acquisition device;
the space-based image acquisition equipment is used for acquiring a space-based orbit image;
the air-based pedestrian boundary intrusion detection device of the rail transit is used for acquiring the air-based rail image from at least one air-based image acquisition device and executing the air-based pedestrian boundary intrusion detection method of the rail transit according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810710554.9A CN109145715B (en) | 2018-07-02 | 2018-07-02 | Air-based pedestrian boundary-crossing detection method, device and system for rail transit |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810710554.9A CN109145715B (en) | 2018-07-02 | 2018-07-02 | Air-based pedestrian boundary-crossing detection method, device and system for rail transit |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109145715A CN109145715A (en) | 2019-01-04 |
| CN109145715B true CN109145715B (en) | 2020-10-30 |
Family
ID=64799662
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810710554.9A Active CN109145715B (en) | 2018-07-02 | 2018-07-02 | Air-based pedestrian boundary-crossing detection method, device and system for rail transit |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109145715B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105759834A (en) * | 2016-03-09 | 2016-07-13 | 中国科学院上海微系统与信息技术研究所 | System and method of actively capturing low altitude small unmanned aerial vehicle |
| CN107204965A (en) * | 2016-03-18 | 2017-09-26 | 阿里巴巴集团控股有限公司 | The hold-up interception method and system of a kind of password cracking behavior |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100801087B1 (en) * | 2006-07-05 | 2008-02-11 | 삼성전자주식회사 | System and method for sensing moving body using structured light, mobile robot including the system |
| JP5831193B2 (en) * | 2011-12-13 | 2015-12-09 | 富士通株式会社 | User detection device, method and program |
| CN106741008B (en) * | 2016-12-29 | 2020-11-06 | 北京交通大学 | Railway line foreign matter identification method and system |
| CN107169401B (en) * | 2017-02-10 | 2020-05-05 | 北京交通大学 | Rail invader detection method based on rail visual feature spectrum |
| CN108009506A (en) * | 2017-12-07 | 2018-05-08 | 平安科技（深圳）有限公司 | Intrusion detection method, application server and computer-readable recording medium |
-
2018
- 2018-07-02 CN CN201810710554.9A patent/CN109145715B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105759834A (en) * | 2016-03-09 | 2016-07-13 | 中国科学院上海微系统与信息技术研究所 | System and method of actively capturing low altitude small unmanned aerial vehicle |
| CN107204965A (en) * | 2016-03-18 | 2017-09-26 | 阿里巴巴集团控股有限公司 | The hold-up interception method and system of a kind of password cracking behavior |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109145715A (en) | 2019-01-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10078790B2 (en) | Systems for generating parking maps and methods thereof | |
| Asha et al. | Vehicle counting for traffic management system using YOLO and correlation filter | |
| US9846946B2 (en) | Objection recognition in a 3D scene | |
| RU2484531C2 (en) | Apparatus for processing video information of security alarm system | |
| CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
| US20130258107A1 (en) | Method of determining parking lot occupancy from digital camera images | |
| CN107273832B (en) | License plate recognition method and system based on integral channel characteristics and convolutional neural network | |
| Peng et al. | Drone-based vacant parking space detection | |
| Chang et al. | Video analytics in smart transportation for the AIC'18 challenge | |
| CN113593250A (en) | Illegal parking detection system based on visual identification | |
| CN111078946A (en) | Bayonet vehicle retrieval method and system based on multi-target regional characteristic aggregation | |
| CN110852179B (en) | Suspicious personnel invasion detection method based on video monitoring platform | |
| CN111967396A (en) | Processing method, device and equipment for obstacle detection and storage medium | |
| CN112434566A (en) | Passenger flow statistical method and device, electronic equipment and storage medium | |
| CN111881984A (en) | Target detection method and device based on deep learning | |
| Dousai et al. | Detecting humans in search and rescue operations based on ensemble learning | |
| CN113012215A (en) | Method, system and equipment for space positioning | |
| CN110636248B (en) | Target tracking method and device | |
| Zhang et al. | Image-based approach for parking-spot detection with occlusion handling | |
| Poostchi et al. | Spatial pyramid context-aware moving vehicle detection and tracking in urban aerial imagery | |
| Mantini et al. | Camera Tampering Detection using Generative Reference Model and Deep Learned Features. | |
| CN116311166A (en) | Traffic obstacle recognition method and device and electronic equipment | |
| Kročka et al. | Extending parking occupancy detection model for night lighting and snowy weather conditions | |
| CN109145715B (en) | Air-based pedestrian boundary-crossing detection method, device and system for rail transit | |
| CN112818837B (en) | Aerial photography vehicle weight recognition method based on attitude correction and difficult sample perception |
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 |