CN113255697A - High-precision high-altitude parabolic detection system and method under complex scene - Google Patents
High-precision high-altitude parabolic detection system and method under complex scene Download PDFInfo
- Publication number
- CN113255697A CN113255697A CN202110609775.9A CN202110609775A CN113255697A CN 113255697 A CN113255697 A CN 113255697A CN 202110609775 A CN202110609775 A CN 202110609775A CN 113255697 A CN113255697 A CN 113255697A
- Authority
- CN
- China
- Prior art keywords
- target
- parabolic
- sequence
- feature
- detection
- 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.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 126
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000012216 screening Methods 0.000 claims abstract description 36
- 238000001914 filtration Methods 0.000 claims abstract description 30
- 238000000605 extraction Methods 0.000 claims abstract description 27
- 238000013528 artificial neural network Methods 0.000 claims abstract description 7
- 239000000203 mixture Substances 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000013527 convolutional neural network Methods 0.000 claims description 6
- 238000003066 decision tree Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 5
- 230000007797 corrosion Effects 0.000 claims description 5
- 238000005260 corrosion Methods 0.000 claims description 5
- 238000002372 labelling Methods 0.000 claims description 5
- 230000000877 morphologic effect Effects 0.000 claims description 5
- 230000003068 static effect Effects 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 abstract description 4
- 238000010801 machine learning Methods 0.000 abstract description 4
- 238000012407 engineering method Methods 0.000 abstract description 2
- 238000013135 deep learning Methods 0.000 abstract 1
- 238000013461 design Methods 0.000 description 6
- 230000007547 defect Effects 0.000 description 5
- 238000004590 computer program Methods 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000003321 amplification Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
Images
Classifications
-
- 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
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a high-precision high-altitude parabolic detection system and a method under a complex scene, wherein the system comprises: firstly, inputting a group of continuous frame image sequences, and detecting a fast moving object in each frame by a moving object detection module; then extracting low-dimensional features and high-dimensional features of each detected target through a feature extraction module, and storing the low-dimensional features and the high-dimensional features in a feature library; inputting the extracted features into a feature matching module for matching frames before and after a target to generate a parabolic sequence; the trajectory prediction and screening module filters an input parabolic sequence, and mainly filters noise; finally, a neural network is learned through the object class filtering module to screen a specific high-altitude parabolic target. The invention integrates an image characteristic engineering method, a front-and-back frame matching algorithm and a machine learning and deep learning algorithm, can detect and position high-altitude parabolas under complex conditions such as rainy days and nights, and is used for recording and tracking accidents.
Description
Technical Field
The invention relates to a high-precision high-altitude parabolic detection system and method in a complex scene, belongs to the technical field of video monitoring and security, and mainly aims at detecting high-altitude parabolic in high floors of shopping malls and residential districts.
Background
Along with the development of urbanization, the high-altitude parabolic behavior becomes more and more a serious and urgent urban problem to be taken into consideration, and the high-altitude parabolic behavior brings great harm and influence to the society. The difficulty of manually checking the high-altitude object from the historical video data is high, the cost is high, and the evidence is difficult to obtain. Most of the existing high-altitude parabolic detection systems use a simpler image detection means, and the false alarm rate and the missing report rate are relatively high, so that the use of the detection system is greatly discounted.
In recent years, the performance and maturity of hardware such as image detection algorithms, cameras, computers and the like are greatly improved, but the technologies and means cannot be applied to the detection of the high-altitude parabolas, particularly the high-altitude parabolas can be accurately detected under the conditions of complex scenes and complex conditions.
In view of the above, a high-precision image detection technique is urgently needed to solve the problems in the prior art.
Patent 1: a high altitude parabolic warning method and device, CN 110853295A. The patent provides a whole set of high-altitude parabolic detection device design, which comprises the steps of camera installation, target intrusion detection, high-altitude parabolic verification, high-altitude parabolic early warning and improvement of detection accuracy rate through a secondary verification method.
Patent 2: a falling object recognition method and system based on image recognition, CN 111488799A. The patent provides a method for identifying falling objects based on image identification, mainly uses a background modeling method to find out a foreground, and uses a density clustering and quadratic equation fitting method to identify the falling objects from the foreground.
Patent 3: a high altitude parabolic monitoring method, CN 111539388A. The patent proposes a complete set of system for detecting and recording parabolas, which uses the image processing method of frame comparison before and after, binarization and the geometric method to detect the position of the parabolas in the image, and combines the position calibrated by the camera to estimate the building room corresponding to the parabolas, and stores all the data in the database system.
Disclosure of Invention
The prior art has the following disadvantages: patent 1: a high altitude parabolic early warning method and device. The main defects of the design are that a Gabor filter is used, multi-scale image analysis and a secondary analysis method cause the detection speed to be slow, and many small targets are filtered after the image is smooth, so that detection omission is easily caused. Patent 2: a falling object identification method and system based on image identification. The main defects of the design are that many designs need to set empirical values, for example, how to judge the conformity of the 'equation coefficients for two measurements', a threshold value needs to be manually set, the universality of the system is reduced, and the latest machine learning method is not combined. Patent 3: a high altitude parabolic monitoring method. The main defects of the design are that a specific and effective parabolic detection model is not provided, and the position of a high-altitude parabolic object in an image is difficult to accurately position by a front-and-back frame comparison and binarization processing method.
The invention aims to overcome the technical defects in the prior art, solve the technical problems and provide a high-precision high-altitude parabolic detection system and method in a complex scene. The algorithm does not need to configure threshold parameters and can achieve the effect of real-time detection
The invention specifically adopts the following technical scheme: high accuracy high altitude parabola detecting system under complicated scene includes:
a moving object detection module to perform: inputting a group of continuous frame image sequences, detecting a foreground target which moves rapidly in each frame image by a background modeling method, and outputting the foreground target to a feature extraction module;
a feature extraction module to perform: receiving foreground targets input by the moving object detection module, extracting low-dimensional features and high-dimensional features of each detected object target, and storing the low-dimensional features and the high-dimensional features in a feature library;
a feature matching module to perform: receiving the low-dimensional features and the high-dimensional features of the feature extraction module, matching the front frame and the rear frame of the object target to generate a sequence conforming to the parabolic track, and outputting the sequence to the track prediction and screening module;
a trajectory prediction and screening module to perform: receiving the sequence which is output by the characteristic matching module and accords with the parabolic track, filtering noise, and outputting the filtered parabolic track sequence to an object type filtering module;
an object class filtering module to perform: receiving the parabolic track sequence output by the track prediction and screening module, and screening a specific high-altitude parabolic target by learning a neural network; the method specifically comprises the following steps: selecting each sequence CiDetection target F with the largest middle areamNot classifying all the detection targets; f is to bemCutting out the corresponding image area, and normalizing to the size of 32 (width) × 32 (height) × 3 (channel number); inputting the image into a designed convolutional neural network for classification, and finally, outputting and representing C by a full connection layeriIs the probability of a high altitude parabola, thereby further filtering out the unsuitable False Positive.
As a preferred embodiment, the moving object detection module specifically executes the following steps: continuous frames captured under a static camera are used for obtaining foreground images through a Gaussian mixture model GMM; amplifying the foreground image through morphological closed operation, and carrying out corrosion operation on a binary image obtained by the Gaussian mixture model GMM to achieve the purpose of amplifying the foreground image; finding all connected regions in the foreground image on the binary image by a method of finding the connected regions, then fitting the found connected regions by using an external rectangle, filtering out some noises only with the size of pixel points, and only keeping the rectangle with a certain area ratio as a detected falling object; and modeling each foreground target detected by background modeling by using a hierarchical clustering model, searching the adjacent targets around the foreground target to classify the targets into one class, marking a label of classified, executing the same operation on the targets which are not classified next time until all the foreground targets are marked with labels of classified, and finally obtaining a clean foreground image.
As a preferred embodiment, the feature extraction module specifically executes the following steps:
firstly, converting a foreground target obtained by detection of the moving object detection module into more abstract features for subsequent interframe matching and feature library construction; projecting a target frame obtained on a binary image onto an original image to obtain a parabola detected under each frame, expanding the width and the height of the detection frame by one time according to the size of the detection frame, normalizing a detection target image frame to the size of 20 (width) multiplied by 20 (height) pixels before extracting features, and eliminating the influence of image dimensions on the features;
extracting geometric features of each target, wherein the geometric features refer to Area and Perimeter Perimeter; on a binary image, the geometric features specifically refer to the area and the perimeter of each filtered connected domain, and the geometric features are saved into a feature library as low-dimensional features;
extracting the texture feature of each target as a high-dimensional feature, wherein the texture feature refers to the LBP feature of 24 dimensions.
As a preferred embodiment, the specific implementation of the feature matching module includes:
setting m groups of parabolic sequence lists C as { C ═ C in the feature library1,C2,...,CmAnd the m groups of parabolic sequence lists C are respectively obtained by detection from a previous N-1 frame image, k targets are detected in a current N frame image through the moving object detection module, and a feature extraction module extracts a feature F ═ F1,F2,...,FkEach sequence CiThe target feature detected last time is set as F'i,i∈[1,m](ii) a Traversing each object F in the feature Fj,j∈[1,k](ii) a Calculating F'iAnd FjWas obtained as the highest score of F'iThereby to move FjClassify into its corresponding CiSequences, which have already been classified, will not receive new detection targets any more, avoiding repeated classification; wherein, F'iAnd FjThe similarity of (c) is calculated by the following formula (1),
wherein C represents the cosine similarity of LBP characteristics, and the calculation formula is formula (2),
wherein,the similarity of the low-dimensional features is represented,smaller values indicate a higher degree of match between the two targets, wa,wlδ is a constant used to normalize the scale between different features, and is calculated as the following equation (3):
if the current parabolic sequence list C is empty, a sequence C is newly established for each feature in the Fi,i∈[1,k]Each detected object FiEntry sequence CiAs sequence CiThe first detected target;
if the target F is found in the matching processjAnd F'iIf the distance is too far, the matching F 'is selected to be abandoned'iSelecting the next target F'i+1Continue to match if at CiIf no suitable sequence can be found for matching, the current detection target F is determinedjAnd automatically discarding the target as an invalid target.
As a preferred embodiment, the trajectory prediction and screening module specifically executes the following steps:
after passing through the feature matching module, a complete set of parabolic sequences C ═ C is generated1,C2,...,CmFor each sequence CiExtracting a 5-dimensional feature: { u _ speed, v _ speed, seq _ length,size _ partial, curve _ score }, u _ speed denotes the sequence CiThe smaller the u _ speed value, the lower C, since the parabola basically does a vertical motioniThe greater the probability of following a parabolic trajectory, the more it is calculated as the following formula (4), wherein X represents CiThe abscissa of each detection target;
v _ speed denotes the sequence CiThe larger the v _ speed value, the larger the average moving speed of the target in the vertical direction, CiThe greater the probability of following a parabolic trajectory, the more it is calculated as the following formula (5), wherein Y represents CiThe ordinate of each detection target;
seq _ length denotes sequence CiThe larger the seq _ length value is, the larger the number of targets contained in (C) is, the feature is used for removing local noiseiThe greater the probability of conforming to a parabolic trajectory;
size _ partial represents sequence CiArea S of the largest detected targetmaxAnd area S of the minimum targetminThe smaller the quotient of size _ partial, CiThe greater the probability of conforming to a parabolic trajectory;
curve _ score is represented by a quadratic curve ax2+by2+ cxy + dx + ey + f ═ 0 fitting CiR2_ score is used to calculate the fit ratio of the quadratic curve, the closer the curve _ score is to 1, CiThe more in line with the quadratic parabolic curve, CiThe greater the probability of following a parabolic trajectory, and conversely, the closer the curve _ score is to 0, CiThe smaller the probability of conforming to a parabolic trajectory;
for the extracted 5-dimensional track features, the collected parabolic sequences are divided into two types by means of labeling: is a true parabolic trajectory with a label of 1; if the track is noise, the label is 0, the track screening problem is converted into a two-classification problem, and then a C4.5 decision tree is trained to predict a new track, so that the aim of screening out an error track or noise is fulfilled.
The invention also provides a high-precision high-altitude parabolic detection method under a complex scene, which comprises the following steps:
a moving object detection step including: inputting a group of continuous frame image sequences, detecting a foreground target which moves rapidly in each frame image by a background modeling method, and outputting the foreground target to a feature extraction step;
a feature extraction step, comprising: receiving the foreground target input in the moving object detection step, extracting low-dimensional features and high-dimensional features of each detected object target, and storing the low-dimensional features and the high-dimensional features in a feature library;
a feature matching step comprising: receiving the low-dimensional features and the high-dimensional features of the feature extraction step, matching the front frame and the rear frame of the object target to generate a sequence conforming to the parabolic track, and outputting the sequence to the track prediction and screening step;
a trajectory prediction and screening step comprising: receiving the sequence which is output by the characteristic matching step and accords with the parabolic track, filtering noise, and outputting the filtered parabolic track sequence to an object type filtering step;
an object class filtering step for performing: receiving the trajectory prediction and screening the parabolic trajectory sequence output by the step, and screening a specific high-altitude parabolic target by learning a neural network; the method specifically comprises the following steps: selecting each sequence CiDetection target F with the largest middle areamSince the largest detection area also means the largest field of view. Considering real-time performance, classifying all detection targets; f is to bemCutting out the corresponding image area, and normalizing to the size of 32 (width) × 32 (height) × 3 (channel number); inputting the image into a designed convolutional neural network for classification, and finally, outputting and representing C by a full connection layeriIs the probability of a high altitude parabola, thereby further filtering out the unsuitable False Positive.
As a preferred embodiment, the moving object detecting step specifically includes: continuous frames captured under a static camera obtain foreground images through a Gaussian mixture model GMM, namely the camera is fixed; because the falling target is usually very small, the foreground image needs to be amplified through morphological closed operation, and the binary image obtained by the Gaussian mixture model GMM is corroded to achieve the purpose of amplifying the foreground image, and the defect is that large noise is generated; finding all connected regions in the foreground image on the binary image by a method of finding the connected regions, then fitting the found connected regions by using an external rectangle, filtering out some noises only with the size of pixel points, and only keeping the rectangle with a certain area ratio as a detected falling object; because the background modeling is easy to have fragmented foreground, the hierarchical clustering model is utilized to model foreground targets detected by each background modeling, the adjacent targets around the foreground targets are searched for and classified into one class, the label is marked as 'classified', the same operation is carried out on the targets which are not classified next time until all the foreground targets are marked as 'classified', and finally a clean foreground image is obtained.
As a preferred embodiment, the feature extraction step specifically includes: firstly, converting a foreground target obtained by detection in the moving object detection step into more abstract features for subsequent interframe matching and feature library construction; and projecting the target frame obtained on the binary image onto an original image to obtain a parabola detected under each frame, and doubling the width and the height of the detection frame according to the size of the detection frame, so as to increase the receptive field and facilitate the extraction of features. Before extracting features, normalizing a detection target picture frame to the size of 20 (width) multiplied by 20 (height) pixels, and eliminating the influence of the image scale on the features;
extracting geometric features of each target, wherein the geometric features refer to Area and Perimeter Perimeter; on a binary image, the geometric features specifically refer to the area and the perimeter of each filtered connected domain, and the geometric features are saved into a feature library as low-dimensional features;
and extracting the texture feature of each target, wherein the texture feature refers to a 24-dimensional LBP feature, and the LBP feature has rotation invariance to the target object falling at high speed.
If the target detected in each frame cannot be in one-to-one correspondence with the target detected in the previous frame, the parabolic sequence cannot be generated, and therefore, as a preferred embodiment, the feature matching step specifically includes:
setting m groups of parabolic sequence lists C as { C ═ C in the feature library1,C2,...,CmAnd the m groups of parabolic sequence lists C are respectively obtained by detection from a previous N-1 frame image, k targets are detected in a current N frame image through the moving object detection module, and a feature F ═ F is extracted through the feature extraction step1,F2,...,FkEach sequence CiThe target feature detected last time is set as F'i,i∈[1,m](ii) a Traversing each object F in the feature Fj,j∈[1,k](ii) a Calculating F'iAnd FjWas obtained as the highest score of F'iThereby to move FjClassify into its corresponding CiSequences, which have already been classified, will not receive new detection targets any more, avoiding repeated classification; wherein, F'iAnd FjThe similarity of (c) is calculated by the following formula (1),
wherein C represents the cosine similarity of LBP characteristics, and the calculation formula is formula (2),
wherein,the similarity of the low-dimensional features is represented,smaller values indicate a higher degree of match between the two targets, wa,wlδ is a constant used to normalize the scale between different features, and is calculated as the following equation (3):
if the current parabolic sequence list C is empty, a sequence C is newly established for each feature in the Fi,i∈[1,k]Each detected object FiEntry sequence CiAs sequence CiThe first detected target;
if the target F is found in the matching processjAnd F'iIf the distance is too far, the matching F 'is selected to be abandoned'iSelecting the next target F'i+1Continue to match if at CiIf no suitable sequence can be found for matching, the current detection target F is determinedjAnd automatically discarding the target as an invalid target.
Many samples belonging to False Positive in the parabolic sequence obtained by the feature matching module, and therefore, as a preferred embodiment, the trajectory predicting and screening step specifically includes:
after passing the feature matching step, a complete set of parabolic sequences C ═ C is generated1,C2,...,CmFor each sequence CiExtracting a 5-dimensional feature: { u _ speed, v _ speed, seq _ length, size _ partial, curved _ score }, u _ speed denotes the sequence CiThe smaller the u _ speed value, the lower C, since the parabola basically does a vertical motioniThe greater the probability of following a parabolic trajectory, the more it is calculated as the following formula (4), wherein X represents CiThe abscissa of each detection target;
v _ speed denotes the sequence CiThe larger the v _ speed value, the larger the average moving speed of the target in the vertical direction, CiThe greater the probability of following a parabolic trajectory, the more it is calculated as the following formula (5), wherein Y represents CiThe ordinate of each detection target;
seq _ length denotes sequence CiThe larger the seq _ length value is, the larger the number of targets contained in (C) is, the feature is used for removing local noiseiThe greater the probability of conforming to a parabolic trajectory;
size _ partial represents sequence CiArea S of the largest detected targetmaxAnd area S of the minimum targetminThe smaller the size _ partial value, the smaller C, since the parabola has a substantially constant size during the falliThe greater the probability of conforming to a parabolic trajectory;
curve _ score is represented by a quadratic curve ax2+by2+ cxy + dx + ey + f ═ 0 fitting CiR2_ score is used to calculate the fit ratio of the quadratic curve, the closer the curve _ score is to 1, CiThe more in line with the quadratic parabolic curve, CiThe greater the probability of following a parabolic trajectory, and conversely, the closer the curve _ score is to 0, CiThe smaller the probability of conforming to a parabolic trajectory;
for the extracted 5-dimensional track features, the collected parabolic sequences are divided into two types by means of labeling: is a true parabolic trajectory with a label of 1; if the track is noise, the label is 0, the track screening problem is converted into a two-classification problem, and then a C4.5 decision tree is trained to predict a new track, so that the aim of screening out an error track or noise is fulfilled.
The invention also proposes an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the steps of the method being implemented when the processor executes the program.
The invention also proposes a storage medium on which a computer program is stored which, when being executed by a processor, carries out the steps of the method.
The invention achieves the following beneficial effects: the invention provides a high-precision high-altitude parabolic detection method and system in a complex scene, which completely avoid the difficulty and uncontrollable property of manually setting detection parameters, and can accurately detect a parabolic track in a complex scene by using a machine learning model and a deep learning model to assist classification and detection; secondly, the method combining the high-dimensional characteristic and the low-dimensional characteristic is used, so that the matching accuracy of the front frame and the rear frame of the small target object moving at high speed is improved, and the actual problem of target tracking loss is avoided; thirdly, the invention designs a set of feature engineering method with supervision and a machine learning method, so that effective tracks can be simply and effectively screened out, and the method is greatly helpful for filtering false alarms in complex scenes.
Drawings
Fig. 1 is a flow chart of a high-precision high-altitude parabolic detection method in a complex scene.
FIG. 2 is a schematic diagram of the topology of the neural network architecture used by the object class filtering module of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
Example 1: the invention provides a high-precision high-altitude parabolic detection system under a complex scene, which comprises:
a moving object detection module to perform: inputting a group of continuous frame image sequences, detecting a foreground target which moves rapidly in each frame image by a background modeling method, and outputting the foreground target to a feature extraction module;
a feature extraction module to perform: receiving foreground targets input by the moving object detection module, extracting low-dimensional features and high-dimensional features of each detected object target, and storing the low-dimensional features and the high-dimensional features in a feature library;
a feature matching module to perform: receiving the low-dimensional features and the high-dimensional features of the feature extraction module, matching the front frame and the rear frame of the object target to generate a sequence conforming to the parabolic track, and outputting the sequence to the track prediction and screening module;
a trajectory prediction and screening module to perform: receiving the sequence which is output by the characteristic matching module and accords with the parabolic track, filtering noise, and outputting the filtered parabolic track sequence to an object type filtering module;
an object class filtering module to perform: receiving the parabolic track sequence output by the track prediction and screening module, and screening a specific high-altitude parabolic target by learning a neural network; the method specifically comprises the following steps: selecting each sequence CiDetection target F with the largest middle areamNot classifying all the detection targets; f is to bemCutting out the corresponding image area, and normalizing to the size of 32 (width) × 32 (height) × 3 (channel number); inputting the image into a designed convolutional neural network for classification, and finally, outputting and representing C by a full connection layeriIs the probability of a high altitude parabola, thereby further filtering out the unsuitable False Positive.
As a preferred embodiment, the moving object detection module specifically executes the following steps: continuous frames captured under a static camera are used for obtaining foreground images through a Gaussian mixture model GMM; amplifying the foreground image through morphological closed operation, and carrying out corrosion operation on a binary image obtained by the Gaussian mixture model GMM to achieve the purpose of amplifying the foreground image; finding all connected regions in the foreground image on the binary image by a method of finding the connected regions, then fitting the found connected regions by using an external rectangle, filtering out some noises only with the size of pixel points, and only keeping the rectangle with a certain area ratio as a detected falling object; and modeling each foreground target detected by background modeling by using a hierarchical clustering model, searching the adjacent targets around the foreground target to classify the targets into one class, marking a label of classified, executing the same operation on the targets which are not classified next time until all the foreground targets are marked with labels of classified, and finally obtaining a clean foreground image.
As a preferred embodiment, the feature extraction module specifically executes the following steps:
firstly, converting a foreground target obtained by detection of the moving object detection module into more abstract features for subsequent interframe matching and feature library construction; projecting a target frame obtained on a binary image onto an original image to obtain a parabola detected under each frame, expanding the width and the height of the detection frame by one time according to the size of the detection frame, normalizing a detection target image frame to the size of 20 (width) multiplied by 20 (height) pixels before extracting features, and eliminating the influence of image dimensions on the features;
extracting geometric features of each target, wherein the geometric features refer to Area and Perimeter Perimeter; on a binary image, the geometric features specifically refer to the area and the perimeter of each filtered connected domain, and the geometric features are saved into a feature library as low-dimensional features;
and extracting the texture feature of each target, wherein the texture feature refers to the LBP feature of 24 dimensions.
As a preferred embodiment, the specific implementation of the feature matching module includes:
setting m groups of parabolic sequence lists C as { C ═ C in the feature library1,C2,...,CmAnd the m groups of parabolic sequence lists C are respectively obtained by detection from a previous N-1 frame image, k targets are detected in a current N frame image through the moving object detection module, and a feature extraction module extracts a feature F ═ F1,F2,...,FkEach sequence CiThe target feature detected last time is set as F'i,i∈[1,m](ii) a Traversing each object F in the feature Fj,j∈[1,k](ii) a Calculating F'iAnd FjWas obtained as the highest score of F'iThereby to move FjClassify into its corresponding CiSequences, which have already been classified, will not receive new detection targets any more, avoiding repeated classification; wherein, F'iAnd FjThe similarity of (c) is calculated by the following formula (1),
wherein C represents the cosine similarity of LBP characteristics, and the calculation formula is formula (2),
wherein,the similarity of the low-dimensional features is represented,smaller values indicate a higher degree of match between the two targets, wa,wlδ is a constant used to normalize the scale between different features, and is calculated as the following equation (3):
if the current parabolic sequence list C is empty, a sequence C is newly established for each feature in the Fi,i∈[1,k]Each detected object FiEntry sequence CiAs sequence CiThe first detected target;
if the target F is found in the matching processjAnd F'iIf the distance is too far, the matching F 'is selected to be abandoned'iSelecting the next target F'i+1Continue to match if at CiIf no suitable sequence can be found for matching, the current detection target F is determinedjAnd automatically discarding the target as an invalid target.
As a preferred embodiment, the trajectory prediction and screening module specifically executes the following steps:
after passing through the feature matching module, a complete set of parabolic sequences C ═ C is generated1,C2,...,CmFor each sequence CiExtracting a 5-dimensional feature: { u _ speed, v _ speed, seq _ length, size _ partial, curved _ score }, u _ speed denotes the sequence CiThe smaller the u _ speed value, the lower C, since the parabola basically does a vertical motioniThe greater the probability of following a parabolic trajectory, the more it is calculated as the following formula (4), wherein X represents CiThe abscissa of each detection target;
v _ speed denotes the sequence CiThe larger the v _ speed value, the larger the average moving speed of the target in the vertical direction, CiThe greater the probability of following a parabolic trajectory, the more it is calculated as the following formula (5), wherein Y represents CiThe ordinate of each detection target;
seq _ length denotes sequence CiThe larger the seq _ length value is, the larger the number of targets contained in (C) is, the feature is used for removing local noiseiThe greater the probability of conforming to a parabolic trajectory;
size _ partial represents sequence CiArea S of the largest detected targetmaxAnd area S of the minimum targetminThe smaller the quotient of size _ partial, CiThe greater the probability of conforming to a parabolic trajectory;
curve _ score is represented by a quadratic curve ax2+by2+ cxy + dx + ey + f ═ 0 fitting CiR2_ score is used to calculate the fit ratio of the quadratic curve, the closer the curve _ score is to 1, CiThe more in line with the quadratic parabolic curve, CiThe greater the probability of following a parabolic trajectory, and conversely, the closer the curve _ score is to 0, CiThe smaller the probability of conforming to a parabolic trajectory;
for the extracted 5-dimensional track features, the collected parabolic sequences are divided into two types by means of labeling: is a true parabolic trajectory with a label of 1; if the track is noise, the label is 0, the track screening problem is converted into a two-classification problem, and then a C4.5 decision tree is trained to predict a new track, so that the aim of screening out an error track or noise is fulfilled.
Example 2: as shown in fig. 1, the present invention further provides a high-precision high-altitude parabolic detection method in a complex scene, specifically, a series of N image sequences is input under a still camera.
Step 1: obtaining a foreground image using a mixed Gaussian model (GMM) starting from the 2 nd image;
step 2: the method mainly carries out corrosion operation on a binary image obtained by GMM through morphologically closed operation amplification prospect to achieve the goal of amplifying a prospect target;
and step 3: finding all connected regions in the image on the binary image, then fitting the connected regions by using an external rectangle to obtain a detection frame, and only keeping the rectangle with a certain area ratio as a detected falling object;
and 4, step 4: modeling each target detected by background modeling by utilizing a hierarchical clustering model, searching the targets in the range of 10 pixels around the target, classifying the targets into one class, marking a label of classified, and executing the same operation on the targets which are not classified next time until the labels of all the targets are classified, and finally obtaining a cleaner foreground;
and 5: and projecting the target frame obtained on the binary image onto the original image to obtain the parabola detected under each frame, wherein the detection frame is respectively doubled in width and height according to the size of the detection frame, so that the receptive field of the target is enlarged. Prior to extracting each target feature, the picture is normalized to a 20 (width) × 20 (height) pixel size;
step 6: and extracting geometric features and textural features of each target, wherein the geometric features comprise the Area size (Area) and the Perimeter (Perimeter) of the region after each connected domain hierarchical clustering. The texture feature refers to a 24-dimensional LBP feature, and the LBP feature has rotation invariance to a target object falling at a high speed;
and 7: suppose that the current 2 nd frame image detects m targets F ═ F through a moving object detection module1,F2,...,FmAnd creating a feature library sequence C, adding each element in F as the first detected element to the sequence C, generating the sequence C ═ C1,X2,...,CmIn which Fi∩Ci=Fi,i∈[1,m],FiThe method comprises the steps of detecting geometric features, textural features and a detection frame of a target;
and 8: repeating the steps 1 to 6 to obtain a group of k targets F ═ F detected on the next frame of image1,F2,...,FkEach sequence CiThe target feature detected last time is set as F'i,i∈[1,m]Go through each target F in Fj,j∈[1,k]Calculating F'iAnd FjTo yield a highest score of F'iThereby to move FjClassification into its corresponding sequence CiC, has been classifiediAnd no new detection target is received any more, and repeated classification is avoided. Wherein F'iAnd FjThe similarity of the two-dimensional image is calculated by the following formula (1);
wherein, C represents the cosine similarity of LBP characteristics, and the calculation formula is as follows;
wherein,representing the similarity of low-dimensional features, with smaller values representing a higher degree of match between two objects, wa=16.0,wl64.0, δ 0.1 is a constant used to normalize the scale between different featuresDegree, the calculation formula is as follows;
if the target F is found in the matching processjDistance F'iToo far away, choose to discard matching F'iSelecting the next target F'i+1Continue to match if at CiIf no suitable sequence can be found for matching, the current detection target F is determinedjBeing an invalid target, is discarded;
and step 9: repeating the step 8 until the N frames of pictures are processed;
step 10: after passing through the feature matching module, a set of matched complete parabolic sequences C ═ C is finally generated1,C2,...,CmFor each sequence CiExtracting a 5-dimensional feature: [ u _ speed, v _ speed, seq _ length, size _ partial, curved _ score](ii) a Step 10-1: calculating a u _ speed characteristic value, wherein the u _ speed represents a sequence CiThe average moving speed of the target in the horizontal direction. The smaller the value, CiThe greater the probability of following a parabolic trajectory, the more the calculation formula is as follows, where X represents CiThe abscissa of each detection target;
step 10-2: calculating v _ speed characteristic value, wherein v _ speed represents a sequence CiAverage moving speed of the target in the vertical direction. The larger the value, CiThe greater the probability of following a parabolic trajectory, the more the calculation formula is as follows, where Y represents CiThe ordinate of each detection target;
step 10-3: calculating the characteristic value of seq _ length, wherein seq _ length represents the sequence CiDetecting the number of targets;
step 10-4: computing the value of the size _ partial feature, size _ partial representing the sequence CiArea S of the largest detected targetmaxAnd area S of the minimum targetminThe quotient of (A) and (B);
step 10-5: the characteristic value of curve _ score is calculated, and curve _ score is represented by a quadratic curve ax2+by2+ cxy + dx + ey + f ═ 0 fitting CiR2_ score is used to calculate the fit ratio of the quadratic curve, the closer the curve _ score is to 1, CiThe more in line with the quadratic parabolic curve, CiThe greater the probability of following a parabolic trajectory, and conversely, the closer the curve _ score is to 0, CiThe smaller the probability of conforming to a parabolic trajectory;
step 10-6: for each extracted sequence CiClassifying by using a trained C4.5 decision tree model, wherein the model input is 5-dimensional features extracted from each sequence, and the output 0 represents CiNot a high-altitude parabolic sequence, 1 represents CiIs a high-altitude parabolic sequence, to obtain the sequence C ' ═ { C ' after filtration '1,C′2,...,C′nWhere C' is a subset of C, n ≦ m;
step 11 (optional): to each sequence C 'in C'iSelecting the detection target F with the largest areamWill FmAn original image corresponding to the detection area is cut out, normalized to the size of a 32 (width) × 32 (height) × 3 (number of channels) image, input to the convolutional neural network designed in fig. 2, and pair C'iAnd (4) performing category judgment, filtering out improper parabolic sequences, and reserving a final detection result.
The invention also proposes an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the steps of the method being implemented when the processor executes the program.
The invention also proposes a storage medium on which a computer program is stored which, when being executed by a processor, carries out the steps of the method.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.
Claims (10)
1. High accuracy high altitude parabola detecting system under complicated scene, its characterized in that includes:
a moving object detection module to perform: inputting a group of continuous frame image sequences, detecting a foreground target which moves rapidly in each frame image by a background modeling method, and outputting the foreground target to a feature extraction module;
a feature extraction module to perform: receiving foreground targets input by the moving object detection module, extracting low-dimensional features and high-dimensional features of each detected object target, and storing the low-dimensional features and the high-dimensional features in a feature library;
a feature matching module to perform: receiving the low-dimensional features and the high-dimensional features of the feature extraction module, matching the front frame and the rear frame of the object target to generate a sequence conforming to the parabolic track, and outputting the sequence to the track prediction and screening module;
a trajectory prediction and screening module to perform: receiving the sequence which is output by the characteristic matching module and accords with the parabolic track, filtering noise, and outputting the filtered parabolic track sequence to an object type filtering module;
an object class filtering module to perform: receiving the parabolic track sequence output by the track prediction and screening module, and screening a specific high-altitude parabolic target by learning a neural network; the method specifically comprises the following steps: selecting each sequence CiDetection target F with the largest middle areamNot classifying all the detection targets; f is to bemCutting out the corresponding image area, and normalizing to the size of 32 (width) × 32 (height) × 3 (channel number); inputting the image into a designed convolutional neural network for classification, and finally, outputting and representing C by a full connection layeriIs the probability of a high altitude parabola, thereby further filtering out the unsuitable False Positive.
2. The high-precision high-altitude parabolic detection system under the complex scene as claimed in claim 1, wherein the moving object detection module specifically executes: continuous frames captured under a static camera are used for obtaining foreground images through a Gaussian mixture model GMM; amplifying the foreground image through morphological closed operation, and carrying out corrosion operation on a binary image obtained by the Gaussian mixture model GMM to achieve the purpose of amplifying the foreground image; finding all connected regions in the foreground image on the binary image by a method of finding the connected regions, then fitting the found connected regions by using an external rectangle, filtering out some noises only with the size of pixel points, and only keeping the rectangle with a certain area ratio as a detected falling object; and modeling each foreground target detected by background modeling by using a hierarchical clustering model, searching the adjacent targets around the foreground target to classify the targets into one class, marking a label of classified, executing the same operation on the targets which are not classified next time until all the foreground targets are marked with labels of classified, and finally obtaining a clean foreground image.
3. The high-precision high-altitude parabolic detection system under the complex scene as claimed in claim 1, wherein the feature extraction module specifically executes:
firstly, converting a foreground target obtained by detection of the moving object detection module into more abstract features for subsequent interframe matching and feature library construction; projecting a target frame obtained on a binary image onto an original image to obtain a parabola detected under each frame, expanding the width and the height of the detection frame by one time according to the size of the detection frame, normalizing a detection target image frame to the size of 20 (width) multiplied by 20 (height) pixels before extracting features, and eliminating the influence of image dimensions on the features;
extracting geometric features of each target, wherein the geometric features refer to Area and Perimeter Perimeter; on a binary image, the geometric features specifically refer to the area and the perimeter of each filtered connected domain, and the geometric features are saved into a feature library as low-dimensional features;
extracting the texture feature of each target as a high-dimensional feature, wherein the texture feature refers to the LBP feature of 24 dimensions.
4. The high-precision high-altitude parabolic detection system under the complex scene as claimed in claim 1, wherein the specific execution of the feature matching module comprises:
setting m groups of parabolic sequence lists C as { C ═ C in the feature library1,C2,...,CmAnd the m groups of parabolic sequence lists C are respectively obtained by detection from a previous N-1 frame image, k targets are detected in a current N frame image through the moving object detection module, and a feature extraction module extracts a feature F ═ F1,F2,...,FkEach sequence CiThe target feature detected last time is set as F'i,i∈[1,m](ii) a Traversing each object F in the feature Fj,j∈[1,k](ii) a Calculating F'iAnd FjWas obtained as the highest score of F'iThereby to move FjClassify into its corresponding CiSequences, which have already been classified, will not receive new detection targets any more, avoiding repeated classification; wherein, F'iAnd FjThe similarity of (c) is calculated by the following formula (1),
wherein C represents the cosine similarity of LBP characteristics, and the calculation formula is formula (2),
wherein,the similarity of the low-dimensional features is represented,smaller values indicate a higher degree of match between the two targets, wa,wlδ is a constant used to normalize the scale between different features, and is calculated as the following equation (3):
if the current parabolic sequence list C is empty, a sequence C is newly established for each feature in the Fi,i∈[1,k]Each detected object FiEntry sequence CiAs sequence CiThe first detected target;
if the target F is found in the matching processjAnd F'iIf the distance is too far, the matching F 'is selected to be abandoned'iSelecting the next target F'i+1Continue to match if at CiIf no suitable sequence can be found for matching, the current detection target F is determinedjAnd automatically discarding the target as an invalid target.
5. The high-precision high-altitude parabolic detection system under the complex scene as claimed in claim 1, wherein the trajectory prediction and screening module is specifically implemented to include:
after passing through the feature matching module, a complete set of parabolic sequences C ═ C is generated1,C2,...,CmFor each sequence CiExtracting a 5-dimensional feature: { u _ speed, v _ speed, seq _ length, size _ partial, curved _ score }, u _ speed denotes the sequence CiThe smaller the u _ speed value, the lower C, since the parabola basically does a vertical motioniThe greater the probability of following a parabolic trajectory, the more it is calculated as the following formula (4), wherein X represents CiThe abscissa of each detection target;
v _ speed denotes the sequence CiThe larger the v _ speed value, the larger the average moving speed of the target in the vertical direction, CiThe greater the probability of following a parabolic trajectory, the more it is calculated as the following formula (5), wherein Y represents CiThe ordinate of each detection target;
seq _ length denotes sequence CiThe larger the seq _ length value is, the larger the number of targets contained in (C) is, the feature is used for removing local noiseiThe greater the probability of conforming to a parabolic trajectory;
size _ partial represents sequence CiArea S of the largest detected targetmaxAnd area S of the minimum targetminThe smaller the quotient of size _ partial, CiThe greater the probability of conforming to a parabolic trajectory;
curve _ score is represented by a quadratic curve ax2+by2+ cxy + dx + ey + f ═ 0 fitting CiR2_ score is used to calculate the fit ratio of the quadratic curve, the closer the curve _ score is to 1, CiThe more in line with the quadratic parabolic curve, CiThe greater the probability of following a parabolic trajectory, and conversely, the closer the curve _ score is to 0, CiThe smaller the probability of conforming to a parabolic trajectory;
for the extracted 5-dimensional track features, the collected parabolic sequences are divided into two types by means of labeling: is a true parabolic trajectory with a label of 1; if the track is noise, the label is 0, the track screening problem is converted into a two-classification problem, and then a C4.5 decision tree is trained to predict a new track, so that the aim of screening out an error track or noise is fulfilled.
6. The high-precision high-altitude parabolic detection method under the complex scene is characterized by comprising the following steps:
a moving object detection step including: inputting a group of continuous frame image sequences, detecting a foreground target which moves rapidly in each frame image by a background modeling method, and outputting the foreground target to a feature extraction step;
a feature extraction step, comprising: receiving the foreground target input in the moving object detection step, extracting low-dimensional features and high-dimensional features of each detected object target, and storing the low-dimensional features and the high-dimensional features in a feature library;
a feature matching step comprising: receiving the low-dimensional features and the high-dimensional features of the feature extraction step, matching the front frame and the rear frame of the object target to generate a sequence conforming to the parabolic track, and outputting the sequence to the track prediction and screening step;
a trajectory prediction and screening step comprising: receiving the sequence which is output by the characteristic matching step and accords with the parabolic track, filtering noise, and outputting the filtered parabolic track sequence to an object type filtering step;
an object class filtering step for performing: receiving the trajectory prediction and screening the parabolic trajectory sequence output by the step, and screening a specific high-altitude parabolic target by learning a neural network; the method specifically comprises the following steps: selecting each sequence CiDetection target F with the largest middle areamNot classifying all the detection targets; f is to bemCutting out the corresponding image area, and normalizing to the size of 32 (width) × 32 (height) × 3 (channel number); inputting the image into a designed convolutional neural network for classification, and finally, outputting and representing C by a full connection layeriIs the probability of a high altitude parabola, thereby further filtering out the unsuitable False Positive.
7. The high-precision high-altitude parabolic detection method under the complex scene according to claim 6, wherein the moving object detection step specifically comprises: continuous frames captured under a static camera are used for obtaining foreground images through a Gaussian mixture model GMM; amplifying the foreground image through morphological closed operation, and carrying out corrosion operation on a binary image obtained by the Gaussian mixture model GMM to achieve the purpose of amplifying the foreground image; finding all connected regions in the foreground image on the binary image by a method of finding the connected regions, then fitting the found connected regions by using an external rectangle, filtering out some noises only with the size of pixel points, and only keeping the rectangle with a certain area ratio as a detected falling object; and modeling each foreground target detected by background modeling by using a hierarchical clustering model, searching the adjacent targets around the foreground target to classify the targets into one class, marking a label of classified, executing the same operation on the targets which are not classified next time until all the foreground targets are marked with labels of classified, and finally obtaining a clean foreground image.
8. The high-precision high-altitude parabolic detection method under the complex scene according to claim 6, wherein the feature extraction step specifically comprises:
firstly, converting a foreground target obtained by detection in the moving object detection step into more abstract features for subsequent interframe matching and feature library construction; projecting a target frame obtained on a binary image onto an original image to obtain a parabola detected under each frame, expanding the width and the height of the detection frame by one time according to the size of the detection frame, normalizing a detection target image frame to the size of 20 (width) multiplied by 20 (height) pixels before extracting features, and eliminating the influence of image dimensions on the features;
extracting geometric features of each target, wherein the geometric features refer to Area and Perimeter Perimeter; on a binary image, the geometric features specifically refer to the area and the perimeter of each filtered connected domain, and the geometric features are saved into a feature library as low-dimensional features;
extracting the texture feature of each target as a high-dimensional feature, wherein the texture feature refers to the LBP feature of 24 dimensions.
9. The high-precision high-altitude parabolic detection method under the complex scene according to claim 6, wherein the feature matching step specifically comprises:
setting m groups of parabolic sequence lists C as { C ═ C in the feature library1,C2,...,CmThe m groups of parabolic sequence lists C are respectively obtained by detection from the previous N-1 frame images, and the current N frame image passes through the moving object detection moduleDetecting k targets, and extracting features F ═ F by the feature extraction step1,F2,...,FkEach sequence CiThe target feature detected last time is set as F'i,i∈[1,m](ii) a Traversing each object F in the feature Fj,j∈[1,k](ii) a Calculating F'iAnd FjWas obtained as the highest score of F'iThereby to move FjClassify into its corresponding CiSequences, which have already been classified, will not receive new detection targets any more, avoiding repeated classification; wherein, F'iAnd FjThe similarity of (c) is calculated by the following formula (1),
wherein C represents the cosine similarity of LBP characteristics, and the calculation formula is formula (2),
wherein,the similarity of the low-dimensional features is represented,smaller values indicate a higher degree of match between the two targets, wa,wlδ is a constant used to normalize the scale between different features, and is calculated as the following equation (3):
if the current parabolic sequence list C is empty, a sequence C is newly established for each feature in the Fi,i∈[1,k]Examine eachMeasured object FiEntry sequence CiAs sequence CiThe first detected target;
if the target F is found in the matching processjAnd F'iIf the distance is too far, the matching F 'is selected to be abandoned'iSelecting the next target F'i+1Continue to match if at CiIf no suitable sequence can be found for matching, the current detection target F is determinedjAnd automatically discarding the target as an invalid target.
10. The high-precision high-altitude parabolic detection method under the complex scene according to claim 6, wherein the track prediction and screening step is specifically executed by:
after passing the feature matching step, a complete set of parabolic sequences C ═ C is generated1,C2,...,CmFor each sequence CiExtracting a 5-dimensional feature: { u _ speed, v _ speed, seq _ length, size _ partial, curved _ score }, u _ speed denotes the sequence CiThe smaller the u _ speed value, the lower C, since the parabola basically does a vertical motioniThe greater the probability of following a parabolic trajectory, the more it is calculated as the following formula (4), wherein X represents CiThe abscissa of each detection target;
v _ speed denotes the sequence CiThe larger the v _ speed value, the larger the average moving speed of the target in the vertical direction, CiThe greater the probability of following a parabolic trajectory, the more it is calculated as the following formula (5), wherein Y represents CiThe ordinate of each detection target;
seq_length represents the sequence CiThe larger the seq _ length value is, the larger the number of targets contained in (C) is, the feature is used for removing local noiseiThe greater the probability of conforming to a parabolic trajectory;
size _ partial represents sequence CiArea S of the largest detected targetmaxAnd area S of the minimum targetminThe smaller the quotient of size _ partial, CiThe greater the probability of conforming to a parabolic trajectory;
curve _ score is represented by a quadratic curve ax2+by2+ cxy + dx + ey + f ═ 0 fitting CiR2_ score is used to calculate the fit ratio of the quadratic curve, the closer the curve _ score is to 1, CiThe more in line with the quadratic parabolic curve, CiThe greater the probability of following a parabolic trajectory, and conversely, the closer the curve _ score is to 0, CiThe smaller the probability of conforming to a parabolic trajectory;
for the extracted 5-dimensional track features, the collected parabolic sequences are divided into two types by means of labeling: is a true parabolic trajectory with a label of 1; if the track is noise, the label is 0, the track screening problem is converted into a two-classification problem, and then a C4.5 decision tree is trained to predict a new track, so that the aim of screening out an error track or noise is fulfilled.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110609775.9A CN113255697B (en) | 2021-06-01 | 2021-06-01 | High-precision high-altitude parabolic detection system and method in complex scene |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110609775.9A CN113255697B (en) | 2021-06-01 | 2021-06-01 | High-precision high-altitude parabolic detection system and method in complex scene |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113255697A true CN113255697A (en) | 2021-08-13 |
CN113255697B CN113255697B (en) | 2024-07-26 |
Family
ID=77185663
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110609775.9A Active CN113255697B (en) | 2021-06-01 | 2021-06-01 | High-precision high-altitude parabolic detection system and method in complex scene |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113255697B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114998794A (en) * | 2022-05-31 | 2022-09-02 | 天翼爱音乐文化科技有限公司 | High-altitude parabolic recognition method, system, device and storage medium |
CN117218517A (en) * | 2023-11-08 | 2023-12-12 | 诺比侃人工智能科技(成都)股份有限公司 | Outdoor moving object detection system in rainy and snowy weather |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190034734A1 (en) * | 2017-07-28 | 2019-01-31 | Qualcomm Incorporated | Object classification using machine learning and object tracking |
CN109309811A (en) * | 2018-08-31 | 2019-02-05 | 中建三局智能技术有限公司 | A kind of throwing object in high sky detection system based on computer vision and method |
CN111079663A (en) * | 2019-12-19 | 2020-04-28 | 深圳云天励飞技术有限公司 | High-altitude parabolic monitoring method and device, electronic equipment and storage medium |
CN111401311A (en) * | 2020-04-09 | 2020-07-10 | 苏州海赛人工智能有限公司 | High-altitude parabolic recognition method based on image detection |
CN112257557A (en) * | 2020-10-20 | 2021-01-22 | 中国电子科技集团公司第五十八研究所 | High-altitude parabolic detection and identification method and system based on machine vision |
CN112818766A (en) * | 2021-01-18 | 2021-05-18 | 深圳英飞拓科技股份有限公司 | High-altitude parabolic detection alarm method and system based on computer vision |
-
2021
- 2021-06-01 CN CN202110609775.9A patent/CN113255697B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190034734A1 (en) * | 2017-07-28 | 2019-01-31 | Qualcomm Incorporated | Object classification using machine learning and object tracking |
CN109309811A (en) * | 2018-08-31 | 2019-02-05 | 中建三局智能技术有限公司 | A kind of throwing object in high sky detection system based on computer vision and method |
CN111079663A (en) * | 2019-12-19 | 2020-04-28 | 深圳云天励飞技术有限公司 | High-altitude parabolic monitoring method and device, electronic equipment and storage medium |
CN111401311A (en) * | 2020-04-09 | 2020-07-10 | 苏州海赛人工智能有限公司 | High-altitude parabolic recognition method based on image detection |
CN112257557A (en) * | 2020-10-20 | 2021-01-22 | 中国电子科技集团公司第五十八研究所 | High-altitude parabolic detection and identification method and system based on machine vision |
CN112818766A (en) * | 2021-01-18 | 2021-05-18 | 深圳英飞拓科技股份有限公司 | High-altitude parabolic detection alarm method and system based on computer vision |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114998794A (en) * | 2022-05-31 | 2022-09-02 | 天翼爱音乐文化科技有限公司 | High-altitude parabolic recognition method, system, device and storage medium |
CN117218517A (en) * | 2023-11-08 | 2023-12-12 | 诺比侃人工智能科技(成都)股份有限公司 | Outdoor moving object detection system in rainy and snowy weather |
CN117218517B (en) * | 2023-11-08 | 2024-01-26 | 诺比侃人工智能科技(成都)股份有限公司 | Outdoor moving object detection system in rainy and snowy weather |
Also Published As
Publication number | Publication date |
---|---|
CN113255697B (en) | 2024-07-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Aboah | A vision-based system for traffic anomaly detection using deep learning and decision trees | |
CN110738127A (en) | Helmet identification method based on unsupervised deep learning neural network algorithm | |
Dedeoğlu et al. | Silhouette-based method for object classification and human action recognition in video | |
CN108491797A (en) | A kind of vehicle image precise search method based on big data | |
Li et al. | Robust people counting in video surveillance: Dataset and system | |
Zhou et al. | Seed-based approach for automated crack detection from pavement images | |
CN113255697B (en) | High-precision high-altitude parabolic detection system and method in complex scene | |
CN111402298A (en) | Grain depot video data compression method based on target detection and trajectory analysis | |
Abidin et al. | A systematic review of machine-vision-based smart parking systems | |
CN113362374A (en) | High-altitude parabolic detection method and system based on target tracking network | |
Tao et al. | Smoky vehicle detection based on multi-scale block Tamura features | |
CN111582166B (en) | Method for detecting remnants based on Gaussian modeling and YoLo V3 target detection | |
Pavani et al. | Comparison of KNN, ANN, CNN and YOLO algorithms for detecting the accurate traffic flow and build an Intelligent Transportation System | |
Yang et al. | Video anomaly detection for surveillance based on effective frame area | |
CN113657250A (en) | Flame detection method and system based on monitoring video | |
CN113627383A (en) | Pedestrian loitering re-identification method for panoramic intelligent security | |
Ng et al. | Vision-based activities recognition by trajectory analysis for parking lot surveillance | |
Kodwani et al. | Automatic license plate recognition in real time videos using visual surveillance techniques | |
CN116453069A (en) | Expressway casting object detection method based on cascade difference perception model | |
Roy | Automatics number plate recognition using convolution neural network | |
Zhao et al. | Research on vehicle detection and vehicle type recognition under cloud computer vision | |
CN114299438A (en) | Tunnel parking event detection method integrating traditional parking detection and neural network | |
Li et al. | Real-time system for tracking and classification of pedestrians and bicycles | |
CN113158720A (en) | Video abstraction method and device based on dual-mode feature and attention mechanism | |
Han et al. | License plate detection and recognition using a dual-camera module in a large space |
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 |