CN-122023942-A - Visual identification method and system based on artificial intelligence

CN122023942ACN 122023942 ACN122023942 ACN 122023942ACN-122023942-A

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a visual identification method and system based on artificial intelligence. The method comprises the steps of obtaining a scene identifier, a visual probe list and installation geometric parameters thereof, generating a session configuration package, carrying out version registration, synchronously collecting image frames of each visual probe, constructing a quality label set and forming an evidence credibility table, executing target detection reasoning and target tracking reasoning, assembling an evidence slot record set according to a task slot template, carrying out cross-probe consistency verification and anti-evidence verification, generating an abnormal conclusion record and a risk classification mark, generating a closed loop feedback record according to a rechecking result, and updating abnormal threshold parameters and slot weight parameters. The invention realizes the structural organization and consistency judgment of the multi-probe visual evidence by configuring the closed loop links with version penetration, evidence admission and evidence countercheck linkage, and improves the stability, traceability and continuous evolution capability of the visual identification process.

Inventors

ZHANG CHUNLIN
LI JUNHONG
YAN SHENGLI
LI HAO
LIU MENG
WANG PENG
Long Huilin
CHEN CHUANG

Assignees

广安职业技术学院
成都远程巨科科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260402

Claims (10)

1. A visual recognition method based on artificial intelligence, comprising: S100, acquiring a scene identifier, a visual probe list, installation geometric parameters of each visual probe, synchronous trigger parameters and task slot template versions, generating a session configuration package and registering session configuration Bao Haxi and version numbers; s200, synchronously acquiring image frames of each visual probe based on a session configuration packet, calculating a definition index, an exposure index, a shielding proportion, a jitter index and a time synchronous deviation, and generating a quality label set; S300, performing target detection reasoning and target tracking reasoning based on an admittance evidence frame set to generate a candidate target set, and assembling the candidate target set into an evidence slot record set according to a task slot template version, wherein the evidence slot record set comprises a probe number, a time window index, a target feature abstract, evidence credibility and a slot gap mark; s400, based on the evidence slot record set, performing cross-probe consistency verification to generate a consistency scoring table, triggering anti-evidence verification on records meeting abnormal threshold conditions in the consistency scoring table, calling the complement acquisition evidence corresponding to the complement acquisition result table through the anti-evidence verification, and carrying out repeated reasoning to output an anti-evidence result table; S500, generating an abnormal conclusion record and a risk classification mark based on the consistency score table and the anti-evidence result table, matching a disposal strategy library version according to the risk classification mark and generating a disposal instruction set, and executing the disposal instruction set to generate a disposal record; S600, generating a closed loop feedback record based on a treatment record access rechecking result, wherein the closed loop feedback record comprises a false alarm mark, a missing report mark and an uncertain mark, updating an abnormal threshold parameter and a slot weight parameter according to the closed loop feedback record, generating a threshold weight updating packet and registering an updating version number.
2. The method of claim 1, wherein the session configuration package comprises a probe role annotation table, a field of view coverage relation table, a unified time reference field, a task slot template version field, and a disposition policy library version field.
3. The method of claim 1, wherein the admission gating employs a hard threshold rule including an upper occlusion ratio threshold and an upper time synchronization deviation threshold, and image frames that do not satisfy the hard threshold rule are written with a to-be-compensated acquisition merge registration slot gap mark.
4. The method of claim 1, wherein the evidence reliability table is generated by a set of quality labels aggregated by weight vectors given by quality weight version fields in a session configuration package.
5. The method of claim 1, wherein the candidate target set comprises target frame coordinates, class labels, confidence scores, appearance feature vectors, and track numbers, wherein the target tracking reasoning uses track numbers to correlate with the appearance feature vectors across frames and outputs track continuity labels.
6. The method of claim 1, wherein the task slot template version defines identity consistency slots, time sequence continuity slots, geometry consistency slots, environment confidence slots, and the evidence slot record set stores slot values, slot source probe numbers, slot source time window indexes, and slot gap marks in terms of slot numbers.
7. The method of claim 1, wherein the cross-probe consistency check generates an identity consistency score, a location continuity score, a geometric constraint residual, an event causal consistency score, and generates a consistency score table according to a score fusion rule.
8. The method of claim 1, wherein the forensic verification comprises a forensic time window index and a forensic probe number, and wherein the forensic verification performs the same target detection reasoning and evidence slot assembly as S300 on the forensic evidence, and wherein the forensic result table comprises a forensic consistency score and a forensic pass flag.
9. The method of claim 1, wherein the treatment instruction set comprises an alarm instruction, a benefit instruction, a parameter adjustment instruction, and a review instruction, the treatment record comprises a treatment policy library version, a evidence slot record set hash, an abnormal conclusion record identification, and a risk classification flag, the threshold weight update package comprises an abnormal threshold parameter update amount and a slot weight parameter update amount, and the abnormal threshold parameter update amount and the slot weight parameter update amount are statistically generated by closed loop feedback records according to false positive marks, false negative marks, and uncertain mark groups.
10. An artificial intelligence based visual recognition system for use in the method of any one of claims 1 to 9, comprising: the configuration and version management module is used for acquiring scene identification, a visual probe list, installation geometric parameters of each visual probe, synchronous triggering parameters, a task slot template version and a treatment strategy library version and generating a session configuration package; The data acquisition and quality evaluation module synchronously acquires image frames of each visual probe based on the session configuration packet, calculates a definition index, an exposure index, a shielding proportion, a jitter index and a time synchronization deviation, and generates a quality label set; the reasoning and slot position assembly module is used for executing target detection reasoning and target tracking reasoning to generate a candidate target set; the consistency verification and anti-verification module is used for executing cross-probe consistency verification and generating a consistency scoring table; The disposal and recording module is used for generating an abnormal conclusion record and a risk grading mark based on the consistency scoring table; and the closed-loop feedback and parameter updating module is used for generating a closed-loop feedback record based on the treatment record access rechecking result.

Description

Visual identification method and system based on artificial intelligence Technical Field The invention relates to the technical field of artificial intelligence, in particular to a visual identification method and system based on artificial intelligence. Background In the field of artificial intelligence technology, existing schemes based on scene identification and visual identification of a visual probe list generally acquire image frames around each visual probe and perform target detection reasoning or target tracking reasoning, and have limitations such as lack of uniform registration caliber of session configuration Bao Haxi and version number, lack of consistent quantitative description of image frame quality difference, lack of structural merging of evidence credibility, and the like. In the existing method, the image frames obtained by multi-reliance single acquisition directly enter reasoning and output, and under the condition that the occlusion proportion, jitter index and time synchronization deviation coexist, the condition that the admission evidence frame set and the collection to be supplemented lack of traceable separation, the quality label set and the evidence credibility table lack of stable association easily occur, and the stable realization of abnormal conclusion record and risk grading mark is difficult to meet. Aiming at the joint processing of a session configuration package, a quality label set, an evidence slot record set, a consistency score table and a evidence-back result table, the prior art has the defects that links such as cross-probe consistency check, abnormal threshold condition triggering, evidence-back verification calling and evidence-back combination corresponding to acquisition to be supplemented and reasoning exist, and the like, and the consistency flow of forming acquisition, admission gating, evidence-assembly slot record set, consistency score table generation, evidence-back result table output, treatment record generation, closed loop feedback record generation and threshold weight update package is difficult in an application scene based on treatment record access check result, so that the update version number of abnormal threshold parameters and slot weight parameters is difficult to form a verifiable closed loop corresponding relation with false report marks, missing report marks and uncertain marks. Disclosure of Invention In order to solve the technical problems, the invention provides a visual identification method based on artificial intelligence, which comprises the following steps: S100, acquiring a scene identifier, a visual probe list, installation geometric parameters of each visual probe, synchronous trigger parameters and task slot template versions, generating a session configuration package and registering session configuration Bao Haxi and version numbers; s200, synchronously acquiring image frames of each visual probe based on a session configuration packet, calculating a definition index, an exposure index, a shielding proportion, a jitter index and a time synchronous deviation, and generating a quality label set; S300, performing target detection reasoning and target tracking reasoning based on an admittance evidence frame set to generate a candidate target set, and assembling the candidate target set into an evidence slot record set according to a task slot template version, wherein the evidence slot record set comprises a probe number, a time window index, a target feature abstract, evidence credibility and a slot gap mark; s400, based on the evidence slot record set, performing cross-probe consistency verification to generate a consistency scoring table, triggering anti-evidence verification on records meeting abnormal threshold conditions in the consistency scoring table, calling the complement acquisition evidence corresponding to the complement acquisition result table through the anti-evidence verification, and carrying out repeated reasoning to output an anti-evidence result table; S500, generating an abnormal conclusion record and a risk classification mark based on the consistency score table and the anti-evidence result table, matching a disposal strategy library version according to the risk classification mark and generating a disposal instruction set, and executing the disposal instruction set to generate a disposal record; S600, generating a closed loop feedback record based on a treatment record access rechecking result, wherein the closed loop feedback record comprises a false alarm mark, a missing report mark and an uncertain mark, updating an abnormal threshold parameter and a slot weight parameter according to the closed loop feedback record, generating a threshold weight updating packet and registering an updating version number. Further, the session configuration package comprises a probe role marking table, a field of view coverage relation table, a unified time reference field, a task slot template version field a