Target video monitoring intelligent tracking method and system based on image recognition

Abstract

The invention discloses a target video monitoring intelligent tracking method and system based on image recognition, and relates to the technical field of target recognition tracking, and the method comprises the steps: collecting monitoring video data, carrying out the marking extraction of a target person needing to be tracked in the monitoring video data, and generating the tracking target person data; performing target person image feature matching analysis on the monitoring video data and the tracking target person data to generate target person image matching analysis data; according to the target video monitoring intelligent tracking method and system based on image recognition, the corresponding target image is searched by marking the target image features according to the target image feature matching, the motion habit of the target is found by collecting the motion features of the target image, the matched target is searched according to the motion habit of the target, and the target video monitoring intelligent tracking efficiency is improved. The method can judge whether the target is a tracking target or not through the motion habit of the target when the appearance characteristics of the target change due to deformation, shielding, illumination change and the like, and further improves the recognition precision.

Description

Technical Field

[0001] The present invention relates to the technical field of target recognition and tracking, and in particular to a target video monitoring intelligent tracking method and system based on image recognition. Background Art

[0002] Video target recognition and tracking technology is a core area of computer vision, and this technology is widely used in security monitoring, autonomous driving, and human-computer interaction. The existing technology with publication number CN119363939A discloses a target tracking system based on video perception, including: a monitoring shooting unit, a data acquisition unit, a data transmission unit, a data storage unit, a data processing unit, a remote monitoring unit, an image correction unit, a feature extraction unit, a matching tracking unit, and a UPS power supply unit; the monitoring shooting unit is used to perform video monitoring shooting through a video shooting device; the data acquisition unit is used to collect data generated during the video perception and tracking of the target in real time, belonging to the field of video tracking technology. By integrating visibility, illumination, light input, noise value, and luminous flux multi-dimensional information acquisition modules, the system can evaluate the shooting environment in real time, automatically adjust image parameters, effectively reduce noise interference, and enhance image clarity, thereby enabling the system to accurately identify target features and achieve fast and accurate target tracking even in complex and changeable low-light environments.

[0003] However, the existing technology still has the problem that the appearance characteristics of the target may change due to deformation, occlusion, lighting changes, etc., which affects the accuracy of target recognition and leads to target tracking errors or loss of tracking targets. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for intelligent tracking of target video surveillance based on image recognition to solve the above-mentioned deficiencies in the prior art.

[0005] In order to achieve the above-mentioned object, the present invention provides the following technical solution: a method for intelligent tracking of target video surveillance based on image recognition, characterized in that:

[0006] S1. Collect surveillance video data, extract tags for the target person to be tracked in the surveillance video data, and generate tracking target person data;

[0007] S2. Performing target person image feature matching analysis on the monitoring video data and the tracking target person data to generate target person image matching analysis data;

[0008] S3, extracting the target person's key point motion features from the target person image matching analysis data to generate target person's key point motion feature data;

[0009] S4. Based on the monitoring video data and the target person's motion feature data, performing target person's motion feature matching analysis to generate target person's key point motion feature matching analysis data;

[0010] S5. Construct target person tracking data, and perform a marking and tracking operation on the target person in the monitoring video data based on the target person tracking data.

[0011] Furthermore, the S1 includes the following steps:

[0012] S11, collect monitoring video data through monitoring cameras, and generate monitoring video data set A=(a1, ..., a w ,…,a υ ), w=1, 2, 3,…, υ, a w represents the wth surveillance video data, and υ represents the maximum number of surveillance video data images;

[0013] S12, select a set number of surveillance video data a w , mark the target person to be tracked, and generate the tracking target person data set B = (b1, ..., b o ,…,b τ ), o=1, 2, 3,..., τ, b o represents the oth tracking target person data, and τ represents the maximum number of tracking target person data.

[0014] Furthermore, the S2 includes the following steps:

[0015] S21, based on the FLANN fast approximate nearest neighbor algorithm, searching the monitoring video data set A for the tracking target person data b in the tracking target person data set B. o Image feature matching target person's surveillance video data a w And the a w Zhong and b o Matching part, generate target person image matching analysis data set C = (c1, ..., c p ,…,c σ ), p=1, 2, 3,…, σ, c p Indicates that the pth sheet exists with b o Matching part of a w and a w Zhong and b o Matching part Right now σ is the maximum number of p.

[0016] Furthermore, the S3 includes the following steps:

[0017] S31, sequentially extracting cp within

[0018] S32, judge the two adjacent Whether the difference between the w numbers is less than the set number threshold;

[0019] S33, if so, then the two Generate a target person's motion feature data d, that is, d = (c p1 ,…,c p2 ), 1≤p1≤p≤p2≤σ, if the target person motion feature data d If the number exceeds the maximum threshold of the number of motion feature images, Start generating the next target person motion feature data d;

[0020] S34, if not, then from the next Start generating the next target person motion feature data d;

[0021] S35, judging the target person's motion feature data d Whether the number is greater than or equal to the minimum threshold of the number of motion feature images, if not, the target person motion feature data d is discarded;

[0022] S36, collecting the target person's motion feature data d, generating a target person's motion feature data set D = (d1, ..., d q ,…,d ω ), q=1, 2, 3,..., ω, d q represents the qth target person motion feature data, ω represents the maximum number of target person motion feature data;

[0023] S37, mark the target person's motion feature data d q Each The key points of the human body are used to generate a key point data set G, where the key points can be the joint positions of the human body;

[0024] S38, the target person's motion feature data d q All The key points of the target person are collected and combined to generate the key point motion feature data E = (e1, ..., e q ,…,e ω ), e q Represents the key point motion feature data of the e-th target person.

[0025] Furthermore, the S4 includes the following steps:

[0026] S41: Detect non-target persons in the surveillance video data set A based on the target detection algorithm, and generate a data set of persons to be identified. i represents the i-th person data set to be identified;

[0027] S42, using the target person motion feature data set D as input and the corresponding key point data set G as output, training a neural network model to generate a key point recognition model;

[0028] S43, the data of the person to be identified is collected Input the key point recognition model to generate the key point data set of the person to be identified Indicates a w Corresponding key point data of the person to be identified;

[0029] S44, in the sliding window with a length equal to the maximum threshold value of the number of motion feature images, in the key point data set of the person to be identified Slide up to generate key point data of the sliding window of the person to be identified

[0030] S45, searching the target person key point motion feature data E for the key point data of the sliding window of the person to be identified based on the particle swarm algorithm. Matching e q , if there is Matching e q Then generate the target person key point motion feature matching analysis data E fenxi , otherwise set K=1, let K=K+1, determine whether K is greater than the set loop threshold, if so, end the loop, otherwise return to S44.

[0031] The particle swarm algorithm is used to search the target person key point motion feature data E for a match with the key point data of the sliding window of the person to be identified, specifically including the following steps:

[0032] a1. Initialize the search particle population parameters, the number of search particle population N, and the maximum number of iterations T;

[0033] a2. Randomly generate N particles in E;

[0034] a3. Calculate the fitness of particles;

[0035] a4. Based on the fitness of the particle, the position of the particle with the best historical fitness is updated to the local optimal position, and the position of the particle with the best fitness among all particles is the global optimal position;

[0036] a5. Particles at the local optimal position G pbest (t) and the global optimal position G gbest (t) under the influence of the movement speed is adjusted to v(t+1), v(t+1)=αv(t)+β(G pbest(t)-G(t))+χ(G gbest (t)-G(t)), where α is the velocity inertia of the search particle, β and χ represent random numbers that balance the local search ability and the global search ability, respectively;

[0037] a6. Determine whether the particle's speed is greater than the set boundary speed v max If so, let v(t+1)=v max ;

[0038] a7. The search particle moves to a new position G(t+1) at v(t+1), G(t+1)=(G(t)+v(t+1));

[0039] a8. Determine whether the number of iterations reaches the maximum number of iterations T. If not, return to step a3. If so, output the global optimal position G. gbest (t) corresponding to e q , if there is Matching e q Then generate the target person key point motion feature matching analysis data E fenxi .

[0040] Furthermore, the S5 includes the following steps:

[0041] S51, matching and analyzing the target person image data set C and the target person key point motion feature matching and analysis data set E. fenxi Collect and combine to generate target person tracking data Z = (C, E fenxi );

[0042] S52, tracking C and E in the target person tracking data Z. fenxi , for C and E in surveillance video data A fenxi Perform tag tracking operations.

[0043] The target video surveillance intelligent tracking system based on image recognition includes a video surveillance module, an input module, a storage, a processor, and a display module;

[0044] The video surveillance module is used to record surveillance videos and generate surveillance video data;

[0045] The input module is used by users to mark the target person to be tracked in the surveillance video data;

[0046] The memory is used to store computer programs;

[0047] The processor is used to execute a computer program to implement the target video surveillance intelligent tracking method based on image recognition according to any one of claims 1 to 6;

[0048] The display module is used for visual display.

[0049] Beneficial effects:

[0050] Compared with the existing technology, the target video surveillance intelligent tracking method and system based on image recognition provided by the present invention searches for the corresponding target image by marking the target image features and matching the target image features, and discovers the target's movement habits by collecting the target image's motion features, and searches for matching targets based on the target's movement habits. When the target's appearance features change due to deformation, occlusion, lighting changes, etc., the target can be judged as a tracking target based on the target's movement habits, thereby further enhancing the recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0052] Figure 1 A diagram of method steps provided by an embodiment of the present invention;

[0053] Figure 2 This is a system structure block diagram provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0055] Example embodiments will be described more fully hereinafter with reference to the accompanying drawings, but the example embodiments may be embodied in different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the scope of this disclosure to those skilled in the art.

[0056] In the absence of conflict, the various embodiments of the present disclosure and the various features therein may be combined with each other.

[0057] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0058] The terms used herein are used only to describe specific embodiments and are not intended to limit the present disclosure. As used herein, the singular forms "a," "an," and "the" are also intended to include the plural forms, unless the context clearly indicates otherwise. It will also be understood that when the terms "comprising" and / or "made of" are used in this specification, the presence of the features, wholes, steps, operations, elements, and / or components is specified, but the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or groups thereof is not excluded.

[0059] The embodiments described herein may be described with reference to plan views and / or cross-sectional views, with the aid of idealized schematic diagrams of the present disclosure. Thus, the example illustrations may be modified based on manufacturing techniques and / or tolerances. Therefore, the embodiments are not limited to the embodiments shown in the accompanying drawings, but include modifications of the configurations formed based on the manufacturing process. Therefore, the regions illustrated in the accompanying drawings are schematic in nature, and the shapes of the regions shown in the drawings illustrate specific shapes of the regions of the elements, but are not intended to be limiting.

[0060] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and will not be interpreted as having an idealized or overly formal meaning unless expressly defined as such herein.

[0061] See also Figure 1-Figure 2 ,Intelligent tracking method for target video surveillance based on image recognition,

[0062] S1. Collect surveillance video data, extract tags for the target person to be tracked in the surveillance video data, and generate tracking target person data, including the following steps:

[0063] S11, collect monitoring video data through monitoring cameras, and generate monitoring video data set A=(a1, ..., a w ,…,a υ ), w=1, 2, 3,…, υ, a w represents the wth surveillance video data, and υ represents the maximum number of surveillance video data images;

[0064] S12, select a set number of surveillance video data a w , mark the target person to be tracked, and generate the tracking target person data set B = (b1, ..., b o ,…,b τ ), o=1, 2, 3,..., τ, b o represents the oth tracking target person data, and τ represents the maximum number of tracking target person data.

[0065] S2. Performing target person image feature matching analysis on the surveillance video data and the tracking target person data to generate target person image matching analysis data, including the following steps:

[0066] S21, based on the FLANN fast approximate nearest neighbor algorithm, search for the existence of tracking target person data b in the monitoring video data set A and the tracking target person data set B. o Image feature matching target person's surveillance video data a w and a w Zhong and b o Matching part, generate target person image matching analysis data set C = (c1, ..., c p ,…,c σ ), p=1, 2, 3,…, σ, c p Indicates that the pth sheet exists with b o Matching part of a w and a w Zhong and b o Matching part Right now σ is the maximum number of p.

[0067] S3, extracting the target person's key point motion features from the target person image matching analysis data to generate the target person's key point motion feature data, including the following steps:

[0068] S31, sequentially extract the c in the target person image matching analysis data set C p within

[0069] S32, judge the two adjacent Whether the difference between the w numbers is less than the set number threshold;

[0070] S33, if so, then the two Generate a target person's motion feature data d, that is, d = (c p1 ,…,c p2 ), 1≤p1≤p≤p2≤σ, if the target person motion feature data d If the number exceeds the maximum threshold of the number of motion feature images, Start generating the next target person motion feature data d;

[0071] S34, if not, then from the next Start generating the next target person motion feature data d;

[0072] S35, judging the target person's motion feature data d Whether the number is greater than or equal to the minimum threshold of the number of motion feature images, if not, the target person motion feature data d is discarded;

[0073] S36. Collect the target person's motion feature data d and generate the target person's motion feature data set D = (d1, ..., d q ,…,d ω ), q=1, 2, 3,..., ω, d q represents the qth target person motion feature data, ω represents the maximum number of target person motion feature data;

[0074] S37, mark the target person's motion feature data d q Each The key points of the human body are used to generate a key point data set G, where the key points can be the joint positions of the human body;

[0075] S38, target person motion feature data d q All The key points of the target person are collected and combined to generate the key point motion feature data E = (e1, ..., e q ,…,e ω ), e q Represents the key point motion feature data of the e-th target person.

[0076] S4, based on the monitoring video data and the target person's motion feature data, performing target person's motion feature matching analysis to generate target person's key point motion feature matching analysis data, including the following steps:

[0077] S41: Detect non-target persons in the surveillance video data set A based on the target detection algorithm, and generate a data set of persons to be identified. i represents the i-th person data set to be identified;

[0078] S42, using the target person motion feature data set D as input and the corresponding key point data set G as output, training a neural network model to generate a key point recognition model;

[0079] S43, collect the data of the person to be identified Input the key point recognition model to generate the key point data set of the person to be identified Indicates a w Corresponding key point data of the person to be identified;

[0080] S44, in a sliding window with a length set to the maximum threshold value of the number of motion feature images, in the key point data set of the person to be identified Slide up to generate key point data of the sliding window of the person to be identified

[0081] S45, based on the particle swarm algorithm, search the target person key point motion feature data E for the sliding window key point data of the person to be identified Matching e q , if there is Matching e q Then generate the target person key point motion feature matching analysis data E fenxi , otherwise set K=1, let K=K+1, determine whether K is greater than the set loop threshold, if so, end the loop, otherwise return to S44.

[0082] The particle swarm algorithm is used to search for key point motion feature data E of the target person that matches the key point data of the sliding window of the person to be identified, which specifically includes the following steps:

[0083] a1. Initialize the search particle population parameters, the number of search particle population N, and the maximum number of iterations T;

[0084] a2. Randomly generate N particles in E;

[0085] a3. Calculate the fitness of particles;

[0086] a4. Based on the fitness of the particle, the position of the particle with the best historical fitness is updated to the local optimal position, and the position of the particle with the best fitness among all particles is the global optimal position;

[0087] a5. Particles at the local optimal position G pbest (t) and the global optimal position G gbest (t) under the influence of the movement speed is adjusted to v(t+1), v(t+1)=αv(t)+β(G pbest (t)-G(t))+χ(G gbest (t)-G(t)), where α is the velocity inertia of the search particle, β and χ represent random numbers that balance the local search ability and the global search ability, respectively;

[0088] a6. Determine whether the particle's speed is greater than the set boundary speed v max If so, let v(t+1)=v max ;

[0089] a7. The search particle moves to a new position G(t+1) at v(t+1), G(t+1)=(G(t)+v(t+1));

[0090] a8. Determine whether the number of iterations reaches the maximum number of iterations T. If not, return to step a3. If so, output the global optimal position G. gbest (t) corresponding to e q , if there is Matching eq Then generate the target person key point motion feature matching analysis data E fenxi .

[0091] S5. Constructing target person tracking data, and performing a marking and tracking operation on the target person in the surveillance video data based on the target person tracking data, including the following steps:

[0092] S51, matching and analyzing the target person image data set C and the target person key point motion feature matching and analysis data set E. fenxi Collect and combine to generate target person tracking data Z = (C, E fenxi );

[0093] S52, based on the target person tracking data Z C and E fenxi , for C and E in surveillance video data A fenxi Perform tag tracking operations.

[0094] The present invention also provides a target video monitoring intelligent tracking system based on image recognition, comprising a video monitoring module, an input module, a storage, a processor, and a display module;

[0095] The video surveillance module is used to record surveillance videos and generate surveillance video data;

[0096] The input module is used by users to mark the target person to be tracked in the surveillance video data;

[0097] The memory is used to store computer programs;

[0098] The processor is used to execute a computer program to implement the target video monitoring intelligent tracking method based on image recognition provided by the present invention;

[0099] The display module is used to visualize the system.

[0100] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. An intelligent tracking method for target video surveillance based on image recognition, characterized by: S1. Collect surveillance video data, extract tags for the target person to be tracked in the surveillance video data, and generate tracking target person data; S2. Performing target person image feature matching analysis on the monitoring video data and the tracking target person data to generate target person image matching analysis data; S3, extracting the target person's key point motion features from the target person image matching analysis data to generate target person's key point motion feature data; S4. Based on the monitoring video data and the target person's motion feature data, performing target person's motion feature matching analysis to generate target person's key point motion feature matching analysis data; S5. Construct target person tracking data, and perform a marking and tracking operation on the target person in the monitoring video data based on the target person tracking data.

2. The method for intelligent tracking of target video surveillance based on image recognition according to claim 1, characterized in that: Said S1 comprises the following steps: S11, collect monitoring video data, generate monitoring video data set A = (a1, ..., a w ,…,a υ ), w=1, 2, 3,…, υ, a w represents the wth surveillance video data, and υ represents the maximum number of surveillance video data images; S12, select a set number of surveillance video data a w , mark the target person to be tracked, and generate the tracking target person data set B = (b1, ..., b o ,…,b τ ), o=1, 2, 3,..., τ, b o represents the oth tracking target person data, and τ represents the maximum number of tracking target person data.

3. The method for intelligent tracking of target video surveillance based on image recognition according to claim 2, characterized in that: The S2 comprises the following steps: S21, based on the FLANN algorithm, searching the monitoring video data set A for the tracking target person data b in the tracking target person data set B. o Image feature matching target person's surveillance video data a w And the a w Zhong and b o Matching part, generate target person image matching analysis data set C = (c1, ..., c p ,…,c σ ), p=1, 2, 3,…, σ, c p Indicates that the pth sheet exists with b o Matching part of a w and a w Zhong and b o Matching part Right now σ is the maximum number of p.

4. The method for intelligent tracking of target video surveillance based on image recognition according to claim 3, characterized in that: The S3 includes the following steps: S31, sequentially extracting c p within S32, judge the two adjacent Whether the difference between the w numbers is less than the set number threshold; S33, if so, then the two Generate a target person's motion feature data d, that is, d = (c p1 ,…,c p2 ), 1≤p1≤p≤p2≤σ, if the target person motion feature data d If the number exceeds the maximum threshold of the number of motion feature images, Start generating the next target person motion feature data d; S34, if not, then from the next Start generating the next target person motion feature data d; S35, judging the target person's motion feature data d Whether the number is greater than or equal to the minimum threshold of the number of motion feature images, if not, the target person motion feature data d is discarded; S36, collecting the target person's motion feature data d, generating a target person's motion feature data set D = (d1, ..., d q ,…,d ω ), q=1, 2, 3,..., ω, d q represents the qth target person motion feature data, ω represents the maximum number of target person motion feature data; S37, mark the target person's motion feature data d q Each The key points of the , generate the key point data set G; S38, the target person's motion feature data d q All The key points of the target person are collected and combined to generate the key point motion feature data E = (e1, ..., e q ,…,e ω ), e q Represents the key point motion feature data of the e-th target person.

5. The method for intelligent tracking of target video surveillance based on image recognition according to claim 4, characterized in that: The S4 comprises the following steps: S41: Detect non-target persons in the surveillance video data set A based on the target detection algorithm, and generate a data set of persons to be identified. i represents the i-th person data set to be identified; S42, using the target person motion feature data set D as input and the corresponding key point data set G as output, training a neural network model to generate a key point recognition model; S43, the data of the person to be identified is collected Input the key point recognition model to generate the key point data set of the person to be identified Indicates a w Corresponding key point data of the person to be identified; S44, in the sliding window with a length equal to the maximum threshold value of the number of motion feature images, in the key point data set of the person to be identified Slide up to generate key point data of the sliding window of the person to be identified S45, searching the target person key point motion feature data E for the key point data of the sliding window of the person to be identified based on the particle swarm algorithm. Matching e q , if there is Matching e q Then generate the target person key point motion feature matching analysis data E fenxi , otherwise set K=1, let K=K+1, determine whether K is greater than the set loop threshold, if so, end the loop, otherwise return to S44.

6. The method for intelligent tracking of target video surveillance based on image recognition according to claim 5, characterized in that: The S5 comprises the following steps: S51, matching and analyzing the target person image data set C and the target person key point motion feature matching and analysis data set E. fenxi Collect and combine to generate target person tracking data Z = (C, E fenxi ); S52, tracking C and E in the target person tracking data Z. fenxi , for C and E in surveillance video data A fenxi Perform tag tracking operations.

7. An intelligent tracking system for target video surveillance based on image recognition, configured to execute the intelligent tracking method for target video surveillance based on image recognition according to any one of claims 1 to 6, characterized in that: It includes video monitoring module, input module, storage, processor and display module; The video surveillance module is used to record surveillance videos and generate surveillance video data; The input module is used by users to mark the target person to be tracked in the surveillance video data; The memory is used to store computer programs; The processor is used to execute a computer program to implement the target video surveillance intelligent tracking method based on image recognition according to any one of claims 1 to 6; The display module is used for visual display.

Images