CN117612109A - Step-by-step detection method for non-specific foreign matter invasion limit in railway running track range - Google Patents
Step-by-step detection method for non-specific foreign matter invasion limit in railway running track range Download PDFInfo
- Publication number
- CN117612109A CN117612109A CN202311683815.XA CN202311683815A CN117612109A CN 117612109 A CN117612109 A CN 117612109A CN 202311683815 A CN202311683815 A CN 202311683815A CN 117612109 A CN117612109 A CN 117612109A
- Authority
- CN
- China
- Prior art keywords
- railway
- detection
- network
- track
- image
- 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.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 92
- 230000009545 invasion Effects 0.000 title claims abstract description 27
- 230000011218 segmentation Effects 0.000 claims abstract description 27
- 238000012544 monitoring process Methods 0.000 claims abstract description 5
- 238000000034 method Methods 0.000 claims description 19
- 230000000694 effects Effects 0.000 claims description 13
- 238000012549 training Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 11
- 230000009466 transformation Effects 0.000 claims description 8
- 238000011897 real-time detection Methods 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 230000009193 crawling Effects 0.000 claims description 2
- 238000013135 deep learning Methods 0.000 claims description 2
- 238000000605 extraction Methods 0.000 claims description 2
- 230000004927 fusion Effects 0.000 claims description 2
- 238000005286 illumination Methods 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 claims description 2
- 238000011176 pooling Methods 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims description 2
- 230000004931 aggregating effect Effects 0.000 claims 1
- 230000000873 masking effect Effects 0.000 claims 1
- 238000013508 migration Methods 0.000 claims 1
- 230000005012 migration Effects 0.000 claims 1
- 230000003014 reinforcing effect Effects 0.000 claims 1
- 241001465754 Metazoa Species 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000004888 barrier function Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 231100000817 safety factor Toxicity 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—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; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Train Traffic Observation, Control, And Security (AREA)
Abstract
The invention discloses a step-by-step detection method for non-specific foreign matter invasion in a railway running track range. Based on the characteristics of railway foreign matter intrusion detection, a YOLO v5 target detection network with high real-time performance and light weight and a PP-Liteseg semantic segmentation network are selected for railway non-specific foreign matter intrusion stepwise detection. Before detection, a railway foreign matter intrusion data set is firstly constructed, a target detection network and a semantic segmentation network are respectively trained by using the data set, a railway monitoring image is input into the trained PP-LiteSeg segmentation network, a track is extracted to divide an intrusion region, an intrusion judging module is utilized to judge whether foreign matter intrusion exists, if the intrusion exists, whether a person or a train exists in the detection image of the network through a YOLO v5 detection network, and if the two targets of the person or the train do not exist, an alarm is output. According to the invention, the detection frequency of the target detection network is reduced by analyzing the result of the network segmentation, so that the purpose of reducing the computational power resources is achieved, and the intrusion of foreign matters in the railway driving range can be detected in real time.
Description
[ technical field ]
The invention relates to the technical field of railway foreign matter intrusion detection, in particular to a step-by-step detection method for non-specific foreign matter intrusion in a railway running track range.
[ background Art ]
With the advent of high-speed railways, railways became one of the common vehicles at home and abroad. Along with the continuous construction and development of the strong traffic countries, the railway lines in China are continuously increased, the operation mileage is also rapidly increased, and under the condition, the guarantee of the safe operation of the railway is a very critical work. The most important is to ensure the safe running of the train on the railway track without derailment accidents. Among safety factors affecting derailment of trains, foreign matters invade the running range of railway tracks, so that the damage caused by collision of running trains with the foreign matters is one of the biggest potential safety hazards. The time of occurrence of the railway foreign matter invasion is random and uncertain, so that real-time detection of the foreign matter invasion in the driving range is a key technical challenge for ensuring the safe operation of the railway.
Many researchers have made extensive studies on the problem of foreign body invasion on railways, and have proposed solutions for systems for detecting and preventing foreign body invasion. The existing method for detecting the invasion of the foreign matters into the peripheral range of the railway mainly comprises contact type foreign matter detection and non-contact type foreign matter detection. The contact type foreign matter detection is generally a method for obtaining foreign matter information by directly or indirectly contacting a sensing element with the foreign matter, such as fiber bragg grating detection and power grid detection; the non-contact type foreign matter detection mainly uses the transmission characteristics of sound waves and electromagnetic waves, so that the information of the foreign matter can be obtained without contacting the foreign matter, such as laser, ultrasonic detection, radar detection, infrared barrier method and video detection method.
In the aspect of railway foreign matter invasion real-time detection, the prior art has the following defects:
(1) For only specific classes of foreign matter
The prior art is generally directed to the detection of specific foreign objects on a railway, such as people, stones, animals, trains, etc. However, the foreign matters appearing in the railway driving range are various and different in characteristics, and large foreign matters such as landslide, debris flow or unusual animals are also seriously harmful to the running of the train, so that it is necessary to detect them in real time.
(2) Undivided infringed regions
The prior art generally detects foreign matter directly on the railway, but this approach is relatively inefficient. Because the monitoring cameras along the railway are generally at a depression angle, the scene near the railway is included, and even the scene on the road near the railway is included. For the foreign matter detection task in the railway perimeter, only the foreign matters which invade the limit range and affect the normal operation of the train are detected, and for the method without dividing the limit invasion area, the detected range is a scene which can be shot by the whole camera, so that the detection burden of the model is increased, and meanwhile, a plurality of false detection conditions exist.
In summary, a detection method capable of detecting all foreign matter invasion and high in real-time performance is urgently needed for foreign matter invasion detection in the railway driving range.
[ summary of the invention ]
The invention provides a step-by-step detection method for the invasion limit of unspecified foreign matters in the railway running track range, which effectively expands the application range of the prior art and improves the efficiency and the precision of the invasion detection of the foreign matters in the railway.
The step-by-step detection method for the invasion of the unspecified foreign matters in the railway running track range is characterized in that the unspecified foreign matters refer to any other object except people and trains, which can influence the running safety of the trains. Since the level crossings on the road in China are now banned and railways are also provided along the railway to block the entry of pedestrians, the people present in the railway limit are generally the staff to overhaul the railway. In this case, the person appearing in the image cannot be counted as a foreign matter that intrudes into the railway limit. The normally running train appearing in the image cannot be counted as foreign matter, except for a person. Aiming at the real-time detection of unspecific foreign matters in the railway running track range, the invention designs a stepwise detection method for the invasion of the unspecific foreign matters of the railway based on a target detection network and a semantic segmentation network.
A stepwise detection method for non-specific foreign matter invasion in a railway running track range is characterized by being capable of detecting foreign matters entering and exiting the invasion track running range in real time and not limiting the types of the foreign matters. According to the method, firstly, a track is extracted through a segmentation network to divide an intrusion region, then an intrusion judging module is utilized to judge whether foreign matter intrusion exists, and finally, a target detection network is utilized to detect whether people or trains exist in an image.
For railway foreign matter intrusion detection, the requirements of real-time performance and light weight are required to be met, so that a YOLO v5s network is selected as a target detection network, and a PP-Liteseg network is selected as a semantic segmentation network. Before railway non-specific foreign object intrusion step-by-step detection is performed using the YOLO v5s target detection network and the PP-Liteseg semantic segmentation network, training needs to be performed on the corresponding railway foreign object intrusion data set so that the model can learn how to identify the characteristics of the target. Because the prior research does not have a disclosed railway foreign matter intrusion data set, the invention establishes a data set with a label by researching the characteristics and the labeling method of railway foreign matter intrusion detection, and specifically comprises the following steps:
step 1: the railway image data acquisition mainly comprises manual acquisition, open source data and network crawling, wherein network data is added in a railway foreign matter intrusion data set, so that the generalization of a network can be improved;
step 2: the invention mainly adopts a manual marking mode, and uses labelme to mark target detection data and semantic segmentation data;
step 3: the railway image data enhancement mode mainly comprises geometric transformation, illumination transformation, mixed transformation and shielding transformation.
After the construction of the railway foreign object intrusion dataset is completed, the dataset is used to train the target detection network and the semantic segmentation network, respectively. The training from scratch requires a great deal of effort and computation, so the method adopts a pre-training weight method based on transfer learning, and most of the pre-training weights are optimal weights obtained by training on a large reference data set, so that the method has better generalization.
The method for extracting the track intrusion region by the PP-Liteseg semantic segmentation network specifically comprises the following steps:
step 1: acquiring railway limit video images;
step 2: extracting feature codes from different network layers;
step 3: the pyramid pooling module aggregates global contexts;
step 4: the decoder fuses detail features and semantic features from high level to low level and uses a unified attention fusion module to augment the feature representation;
step 5: the mask outputs the track portion.
And evaluating the extraction effect of the PP-Liteseg model on the railway track intrusion region based on the average cross-over ratio mIoU, the pixel accuracy PA and the Dice coefficient.
The mlou is the sum of the intersection ratios of all the categories and then the average value is obtained, and considering that the invention only relates to the track category, the part except the track in the image is identified as the background, so that the mlou is equal to the track intersection ratio IoU.
PA is the ratio of the predicted correct number of pixels to all pixels in the image, and the expression is calculated as follows:
where N is the total number of categories, N ii Refers to the number of pixels, m, of which the i-th class is correctly predicted as the i-th class i Is the total number of pixels of the i-th class.
The Dice coefficient is used to measure the similarity between two sets, and the expression can be derived from the confusion matrix:
where X and Y are two sets, TP is a true case, FP is a false case, and FN is a false counterexample.
The intrusion judging module is used for judging whether the track part in the image is blocked or not, entering the next target detection network if the track part is blocked, otherwise, not entering the subsequent detection network if the track part is not blocked. And judging whether the track duty ratio difference exceeds a threshold value or not, and judging whether the target detection network is entered or not. The method specifically comprises the following steps:
step 1: calculating the duty ratio of a track part in the current frame image segmented by the semantic segmentation network in the whole image;
step 2: adaptively replacing a maximum track occupancy ratio;
step 3: subtracting the track occupation ratio of the current frame image from the maximum track occupation ratio to obtain an absolute value;
step 4: judging whether the absolute value exceeds a set threshold value, if so, indicating that the current frame image has an object, entering a next step 5, otherwise, not allowing the object to invade the track area, and returning to the step 1;
step 5: the current frame image is input to the object detection network.
The YOLO v5s target detection network is used for detecting whether two targets of a person or a train exist in an input image, if the two targets do not exist, an alarm is sent to a corresponding dispatching command control center, and staff timely react to avoid the phenomenon that the train collides with foreign matters to derail. Otherwise, if a person or train is present in the input image, no alarm is selected. The method specifically comprises the following steps:
step 1: inputting a current frame image with the track occupation ratio difference exceeding a set threshold;
step 2: preprocessing an input image, and adopting an adaptive calculation anchor frame and an adaptive image scaling method;
step 3: the method comprises the steps of slicing an original image with the size of 640 multiplied by 3 through a backbone network consisting of a Focus network structure and a CSP network structure to obtain a feature map with the size of 320 multiplied by 12, and performing convolution operation through 32 convolution kernels to obtain the feature map with the size of 320 multiplied by 12;
step 4: through adopting a Neck network with an FPN+PAN structure;
step 5: outputting a target identification result, if no two targets of a person or a train exist, entering a step 6, otherwise, returning to the step 1;
step 6: an alarm is output.
And evaluating the detection effect of the YOLO v5s model on people and trains within the railway perimeter range based on the precision P, the recall R, the intersection ratio IoU and the average mean precision mAP.
The precision P is used for representing the precision of the final detection result, and the calculation formula of the precision P is as follows, wherein the prediction area output by the model is the correct area or not:
the recall rate R is used for representing the comprehensive degree of the model detection result, whether all marked targets in the image are predicted or not is judged, and the calculation formula is as follows:
the blending ratio IoU is used for evaluating the performance of the target detection algorithm, wherein A represents a prediction boundary box, B represents a real boundary box, and the calculation formula of IoU is as follows:
based on the deep learning experience, 0.5 was taken as the threshold for IoU. In the invention, the IoU of the model prediction result is larger than the threshold value of 0.5, and the model prediction result is judged to be a positive sample, otherwise, the model prediction result is judged to be a negative sample.
The average value average precision mAP is used for representing the average value of average precision AP of people and trains, and 0.5 is taken as a IoU threshold value and recorded as [email protected]; the AP is the average accuracy of detecting a person or a train, and the P-R curve can be obtained by calculating P and R of the person or the train during detection, and the size of the area surrounded by the P-R curve and the coordinate system is the value of the AP.
[ advantageous effects ]
The invention provides a step-by-step detection method for non-specific foreign matter invasion in a railway running track range, which can save a large amount of target detection time, and can carry out target detection on an image only when foreign matters possibly exist, so that calculation force resources are greatly reduced, and the detection efficiency of the non-specific foreign matter invasion of a railway is greatly improved. And the existing target detection algorithm and semantic segmentation algorithm are compared and analyzed, the selected YOLO v5s target detection algorithm and the PP-Liteseg semantic segmentation algorithm are high in instantaneity and accuracy, and the two algorithm models are lightweight networks, have lower requirements on hardware, and are more suitable for being deployed on equipment.
[ description of the drawings ]
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a graph showing the effect of track segmentation of a trained PP-LiteSeg model according to an embodiment of the invention;
FIG. 3 is a graph showing the effect of target detection on a trained YOLO v5s model in accordance with an embodiment of the present invention.
Detailed description of the preferred embodiments
The present invention will be further described with reference to the following examples and drawings in order to make the objects, technical solutions and features of the present invention more clear. The examples are only for explaining the present invention and are not limiting thereof.
Step 1: the railway image dataset is constructed and processed, the embodiment collects data at monitoring equipment of key positions along Hukun high-speed rail and Yuhui railways, around Huai high-speed rail stations and near railway stations, 3000 images are provided for training a target detection network through image data labeling and data enhancement, 2000 images are provided for a semantic segmentation network, and the method comprises the following steps: 2, dividing the training set and the verification set in proportion;
step 2: using Python3.7+PaddlePaddle framework development, selecting YOLO v5s target detection algorithm and PP-Liteseg semantic segmentation algorithm, and setting hardware configuration and adopted parameters as follows:
TABLE 1 hardware configuration
Table 2 target detection network parameter settings
Table 3 semantic segmentation network parameter settings
Step 3: training the target detection network and the semantic segmentation network according to the data set, evaluating the training results of the target detection network from three indexes of P, R and [email protected], and evaluating the training results of the semantic segmentation network from three indexes of mIoU, PA and Dice, wherein the detection results are shown in the following table:
TABLE 4Yolo v5s detection results
TABLE 5PP-LiteSeg detection results
The training results show that the MIoU of the PP-LiteSeg segmented track reaches 87%, the detection precision [email protected] of the YOLO v5s on the human and train reaches 91%, and the results show that the two networks are high in detection precision, the two networks are lightweight networks, the occupied calculation force resources are small, and the method is very suitable for being deployed on mobile equipment for detection.
Step 4: inputting a railway monitoring image into a trained PP-LiteSeg model, and performing track segmentation, wherein a segmentation effect diagram is shown in figure 2;
step 5: entering a judging module, calculating the ratio of the track part in the current frame image to the whole image, adaptively changing the maximum track ratio, subtracting the track ratio of the current frame image from the maximum track ratio to obtain an absolute value, judging whether the absolute value exceeds a set threshold, entering the next step 6 if the absolute value exceeds the set threshold, and returning to the step 4 if the absolute value exceeds the set threshold;
step 6: inputting the image into a trained YOLO v5s model for target detection, wherein a target detection effect diagram is shown in fig. 3:
step 7: judging the target recognition result, if no two targets of a person or a train exist, entering a step 8, otherwise, returning to the step 4;
step 8: an alarm is output.
As can be seen from fig. 2, the segmented track edges of the PP-LiteSeg model are not only continuous but also very accurate, and the PP-LiteSeg model does not erroneously recognize background pixels similar to the track as a track, which all indicate that the detection effect of the PP-LiteSeg model corresponds to the detection effect of the track portion required by the present invention. Only if the track is accurately divided, namely, the dangerous area in the periphery of the railway, whether foreign matters invade the dangerous area can be accurately judged, so that the influence of the invasion of the foreign matters on the normal operation of the railway is effectively prevented.
As can be seen from fig. 3, the rectangular frames in the figure are respectively marked with the category of the framed object and the confidence, the confidence refers to the confidence that the model considers that the detection frame is of a certain object type, and from the view of the detection effect graph, the detection effect of people and trains in the image with the railway background is good, and basically no condition of missed detection or false detection occurs.
The step-by-step detection method for the unspecified foreign matter invasion in the railway running track range provided by the invention adopts the lightweight network PP-LiteSeg and YOLO v5s, so that the method is easier to be deployed on mobile equipment, greatly reduces the number of target detection times, greatly saves calculation force resources, and can realize real-time detection of the foreign matter invasion in the railway running range.
Claims (1)
1. A step-by-step detection method for unspecified foreign matter invasion in a railway running track range is characterized by comprising the following steps:
step 1: establishing a railway foreign matter intrusion data set with a label: railway image data is collected through three modes of manual collection, open source data and network crawling, then target detection data and semantic segmentation data are manually marked by labelme, and finally image data enhancement is carried out through methods such as geometric transformation, illumination transformation, mixed transformation and shielding transformation;
step 2: training a target detection network and a semantic segmentation network respectively using the railway foreign object invasion data set: according to the characteristics of railway foreign matter invasion, selecting YOLO v5s and PP-Liteseg as target detection and semantic segmentation networks respectively, and performing network training by adopting a migration learning method;
step 3: extracting a track intrusion region through a PP-Liteseg semantic segmentation network: inputting railway monitoring images into a trained PP-LiteSeg model, extracting feature codes from different network layers, aggregating global contexts through a pyramid pooling module, fusing detail features and semantic features from a high level to a low level by a decoder, reinforcing feature representation by using a unified attention fusion module, outputting a track part by masking, and evaluating the extraction effect of the PP-LiteSeg model on a railway track intrusion region based on an average cross-over ratio mIoU, a pixel accuracy PA and a Dice coefficient;
the mIoU is the average value after summation of the cross-over ratios of all the categories, and considering that the invention only relates to the track category, the rest of the image is identified as background, so the mIoU is equal to the track cross-over ratio IoU;
PA is the ratio of the predicted correct number of pixels to all pixels in the image, specifically:
wherein N is the total number of categories, N ii Correctly predicted as the number of pixels of the ith class for the ith class, m i Is the total number of pixels of the i-th class;
the Dice coefficient is used for measuring the similarity between two sets, and the expression is obtained through the confusion matrix:
wherein X and Y are two sets, TP is a true case, FP is a false case, and FN is a false counterexample;
step 4: the intrusion limit judging module judges whether a track part in the image is shielded or not: firstly, calculating the ratio of a track part in a current frame image segmented by a semantic segmentation network in the whole image, and adaptively replacing the maximum track ratio; subtracting the track occupation ratio of the current frame image from the maximum track occupation ratio to obtain an absolute value, judging whether the absolute value exceeds a set threshold value, if so, indicating that the current frame image has an object, entering a step 5, otherwise, not allowing the object to invade a track area, and returning to the step 3;
step 5: detecting objects in the image through a YOLO v5s object detection network: inputting a current frame image with the track ratio difference exceeding a set threshold value into a trained YOLO v5s model, and preprocessing the input image by adopting an adaptive calculation anchor frame and adaptive image scaling method; then, through a backbone network consisting of a Focus network structure and a CSP network structure, slicing an original image with the size of 640 multiplied by 3 to obtain a feature image with the size of 320 multiplied by 12, then, carrying out convolution operation on 32 convolution kernels to obtain a feature image with the size of 320 multiplied by 12, finally, outputting a target identification result through a Neck network of an FPN+PAN structure, and evaluating the detection effect of a YOLO v5s model on people and trains within the railway perimeter range based on the precision P, the recall rate R, the cross-over ratio IoU and the average precision mAP; judging the identification result, if no person or train exists, entering a step 6, otherwise, returning to the step 3;
the precision P is used for representing the precision of the final detection result, and the calculation formula is as follows:
the recall rate R is used for representing the comprehensive degree of the model detection result, and the calculation formula is as follows:
the intersection ratio IoU is used for evaluating the performance of the target detection algorithm, wherein a represents a prediction boundary box, B represents a real boundary box, and the calculation formula of IoU is as follows:
according to the experience of deep learning, 0.5 is taken as a threshold value of IoU, in the invention, ioU of a model prediction result is larger than the threshold value of 0.5, and the model prediction result is judged to be a positive sample, otherwise, the model prediction result is judged to be a negative sample;
the average value average precision mAP is used for representing the average value of average precision AP of people and trains, and 0.5 is taken as a IoU threshold value and recorded as [email protected]; the AP is the average accuracy of detecting a person or a train, and during detection, a P-R curve can be obtained by calculating P and R of the person or the train, and the size of an area surrounded by the P-R curve and a coordinate system is the value of the AP;
step 6: and outputting an alarm, sending the alarm to a corresponding dispatching command control center, and timely reacting by staff to avoid the phenomenon that the train collides with foreign matters to derail, thereby realizing real-time detection of foreign matter invasion in the railway driving range.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311683815.XA CN117612109A (en) | 2023-12-08 | 2023-12-08 | Step-by-step detection method for non-specific foreign matter invasion limit in railway running track range |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311683815.XA CN117612109A (en) | 2023-12-08 | 2023-12-08 | Step-by-step detection method for non-specific foreign matter invasion limit in railway running track range |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117612109A true CN117612109A (en) | 2024-02-27 |
Family
ID=89953464
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311683815.XA Pending CN117612109A (en) | 2023-12-08 | 2023-12-08 | Step-by-step detection method for non-specific foreign matter invasion limit in railway running track range |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117612109A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118182572A (en) * | 2024-05-16 | 2024-06-14 | 永鼎行远(南京)信息科技有限公司 | Anti-collision early warning device for railway mobile equipment |
-
2023
- 2023-12-08 CN CN202311683815.XA patent/CN117612109A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118182572A (en) * | 2024-05-16 | 2024-06-14 | 永鼎行远(南京)信息科技有限公司 | Anti-collision early warning device for railway mobile equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105809679B (en) | Mountain railway side slope rockfall detection method based on visual analysis | |
| CN109948416A (en) | A kind of illegal occupancy bus zone automatic auditing method based on deep learning | |
| CN109637151A (en) | A kind of recognition methods that highway Emergency Vehicle Lane is driven against traffic regulations | |
| CN111091110B (en) | Reflection vest wearing recognition method based on artificial intelligence | |
| CN107301777A (en) | Vehicle peccancy lane change detection method based on video detection technology | |
| CN109766746A (en) | A kind of track foreign matter detecting method of unmanned plane video | |
| CN117612109A (en) | Step-by-step detection method for non-specific foreign matter invasion limit in railway running track range | |
| CN113553916B (en) | Orbit dangerous area obstacle detection method based on convolutional neural network | |
| CN111667655A (en) | Infrared image-based high-speed railway safety area intrusion alarm device and method | |
| CN112084928A (en) | Road traffic accident detection method based on visual attention mechanism and ConvLSTM network | |
| CN112381014A (en) | Illegal parking vehicle detection and management method and system based on urban road | |
| CN114248819B (en) | Railway intrusion foreign matter unmanned aerial vehicle detection method, device and system based on deep learning | |
| CN113011338A (en) | Lane line detection method and system | |
| CN115995056A (en) | Automatic bridge disease identification method based on deep learning | |
| CN110838230A (en) | Mobile video monitoring method, monitoring center and system | |
| CN111582084A (en) | Weak supervision learning-based rail foreign matter detection method and system under empty base view angle | |
| CN115410114A (en) | Urban rail flood prevention early warning method and system based on multiple characteristics | |
| CN115546742A (en) | Rail foreign matter identification method and system based on monocular thermal infrared camera | |
| CN112861762B (en) | Railway crossing abnormal event detection method and system based on generation countermeasure network | |
| CN106372566A (en) | Digital signage-based emergency evacuation system and method | |
| CN114758414A (en) | Pedestrian behavior detection method, device, equipment and computer storage medium | |
| CN112382068B (en) | Station waiting line crossing detection system based on BIM and DNN | |
| CN116311071A (en) | Substation perimeter foreign matter identification method and system integrating frame difference and CA | |
| CN116311166A (en) | Traffic obstacle recognition method and device and electronic equipment | |
| CN115601733A (en) | Human body skeleton-based method and system for detecting cheating behaviors of three-subject security officer |
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 |