CN109034844A - A kind of food safety trace back system retrieval method based on camera lens sample searching - Google Patents

Abstract

The food safety trace back system retrieval method based on camera lens sample searching that the invention discloses a kind of, specifically include that A. obtains the food camera lens picture for needing to retrieve, shot segmentation extracts key frame, and the feature space that can sufficiently reflect camera lens content is formed according to the set of characteristic points of key frame；B. connection features point forming curves carry out linear approximation to curve, determine most preferably approaching as key frame for the point of inflexion；C. it is determined by the threshold value of characteristic point line and retrieves image at a distance from feature lines all in sample camera lens, and it is ranked up, obtain most like camera lens, complete the retrieval of food safety trace back system.This method has preferable timing, versatility and correlation, improves retrieval precision by approaching extraction key frame combination space length, complexity is reduced, the time required to reducing retrieval.

Description

Food safety tracing system retrieval method based on lens sample retrieval

Technical Field

The invention relates to a retrieval method of a food safety tracing system, and belongs to the fields of food safety, video retrieval and differential geometry.

Background

In recent years, because food safety crisis frequently generated, the health of people is seriously affected, the worldwide attention is attracted, and how to effectively track and trace food becomes an extremely urgent global subject. The food safety tracing system can connect all links of production, inspection, supervision and consumption, so that consumers can know the production and circulation processes which accord with sanitary safety. The system provides a tracing mode from farmland to dining table, extracts public tracing elements concerned by consumers in supply chain links such as production, processing, circulation and consumption, establishes a food safety information database, can trace back to the source according to video retrieval once problems are found, and guarantees the legal rights and interests of the consumers from the source. The existing video retrieval ignores the time sequence relation and the correlation among key frames, the retrieval precision is not accurate enough, and the time is long.

Disclosure of Invention

In order to solve the above problems, the present invention aims to provide a food safety traceability system retrieval method with good timing sequence, universality and relevance, which improves the retrieval accuracy by extracting key frames by approximation in combination with spatial distance, reduces the complexity and reduces the time required for retrieval.

The technical scheme adopted by the invention for solving the problems comprises the following steps:

A. acquiring a food shot picture to be retrieved, segmenting a shot to extract a key frame, and forming a feature space capable of fully reflecting the content of the shot according to a feature point set of the key frame;

B. connecting the characteristic points to form a curve, performing linear approximation on the curve, and determining the optimal approximation of the points of inflection as a key frame;

C. and determining the distances between the retrieval image and all the characteristic connecting lines in the sample lens through the threshold values of the characteristic point connecting lines, and sequencing the distances to obtain the most similar lens so as to finish the retrieval of the food safety tracing system.

Further, the step a comprises:

(1) acquiring food shot pictures to be searched, dividing shots, extracting key frames, finding out a boundary between two continuously appearing shots through detecting shot switching points, gathering the frames belonging to the same shot together, and gathering feature points in each frame to form a feature space capable of fully reflecting the content of the shot;

(2) two key frames F are selected to represent the main content of the shot_a、F_bCorresponding to two feature points f_a、f_b，Is provided with

Where d is the dimension of the feature space, the similarity between two frames is measured by the euclidean distance:

D(f_a,f_b)＝∥f_a-f_b∥。

further, the step B includes:

(1) n feature points in the shot feature set S, structureA connecting line, if there are Q sets in the space, the total characteristic line isProjection of search point r on feature connecting line

f_r＝f_a+k(f_a-f_b)

Wherein,therefore, the similarity between the search point and the shot is:

D(f_r,(f_a,f_b))＝∥r-f_r∥

(3) performing linear approximation on the characteristic point connecting curve by using an approximation line l_al_bApproximation curve f_af_bThe feature space is composed of n-1 feature lines, the line of continuous reverse curve points is used for approaching the feature curve, and the frame corresponding to the reverse curve points is used as the key frame of the lens; setting a threshold value ε, if D (f)_j,f_j+1)>ε or D (f)_j-₁,f_j+1)>c is e, then f_j+1Is a new key frame, where 0<j is n-1, c is a constant and 1<c<2, searching key frames in each feature space, determining the number of key frames in the shot, and determining the number of key frames

The object can reflect the characteristics of the shot, and if no key frame exists, the whole shot can be regarded as a characteristic point; if the number of the key frames is large, the shot content is changed greatly or the set threshold is small, the threshold can be adjusted for recalculation, and if the feature points of the new key frame are on a straight line, the middle redundant feature points can be removed, so that the calculation complexity is reduced.

Further, the step C includes:

(1) setting a threshold value delta of the distance between the connection lines of the feature points, wherein/f only when the distance between the search point and the connection line of the feature points is less than or equal to the threshold value_r-f_j/. ltoreq.δ, the result is accepted, otherwise the characteristic point is not considered to belong to the set;

(2) the method comprises the steps of searching and applying a shot sample to a food safety tracing system, establishing a video library containing a plurality of shots, recording all information including transportation, packaging, subpackaging and sales circulation processes from four links of food production, processing, circulation and consumption, giving food shots needing to be inquired, and displaying an inquiry result according to a distance.

The invention has the beneficial effects that:

under the conditions of large video storage data volume and long storage time, the method has better time sequence, universality and correlation, improves the retrieval precision by extracting key frames by approximation and combining with the space distance, reduces the complexity and reduces the time required by retrieval.

Drawings

FIG. 1 is an overall flow chart of a retrieval method of a food safety tracing system based on lens sample retrieval;

FIG. 2 is a flowchart of an approximation algorithm for extracting key frames;

FIG. 3 is a graph of a food sample query result;

Detailed Description

Referring to fig. 1, the method of the present invention comprises the steps of:

A. acquiring a food shot picture to be retrieved, segmenting a shot to extract a key frame, and forming a feature space capable of fully reflecting the content of the shot according to a feature point set of the key frame;

(1) the method comprises the steps of obtaining food shot pictures needing to be searched, dividing shots, extracting key frames, finding out the boundary between two continuously appearing shots through detecting shot switching points, gathering the frames belonging to the same shot together, and gathering feature points in each frame to form a feature space capable of fully reflecting the content of the shot.

(2) Two key frames F are selected to represent the main content of the shot_a、F_bCorresponding to two feature points f_a、f_b，Is provided with

Where d is the dimension of the feature space. The similarity between two frames is measured by the euclidean distance:

D(f_a,f_b)＝∥f_a-f_b∥

B. connecting the characteristic points to form a curve, performing linear approximation on the curve, and determining the optimal approximation of the point of inflection as a key frame (as shown in FIG. 2);

(1) n feature points in the shot feature set S, structureA connecting line, if there are Q sets in the space, the total characteristic line isProjection of search point r on feature connecting line

f_r＝f_a+k(f_a-f_b)

Wherein,therefore, the similarity between the search point and the shot is:

D(f_r,(f_a,f_b))＝∥r-f_r∥

(2) performing linear approximation on the characteristic point connecting curve by using an approximation line l_al_bApproximation curve f_af_b. The feature space is composed of n-1 feature lines. And (5) approximating the characteristic curve by using a connecting line of continuous reverse curve points, and taking a frame corresponding to the reverse curve points as a key frame of the lens. Setting a threshold value ε, if D (f)_j,f_j+1)>ε or D (f)_j-₁,f_j+1)>c is e, then f_j+1Is a new key frame, where 0<j is n-1, c is a constant and 1<c<2. And finding key frames in each feature space, and determining the number of the key frames in the shot. Number of key frames

The object can reflect the characteristics of the shot, and if no key frame exists, the whole shot can be regarded as a characteristic point; if the number of the key frames is large, the shot content is changed greatly or the set threshold is small, and the threshold can be adjusted for recalculation. If the feature points of the new key frame are on a straight line, the middle redundant feature points can be removed, and the calculation complexity is reduced.

C. And determining the distances between the retrieval image and all the characteristic connecting lines in the sample lens through the threshold values of the characteristic point connecting lines, and sequencing the distances to obtain the most similar lens so as to finish the retrieval of the food safety tracing system.

(1) Setting a threshold value delta of the distance between the connection lines of the feature points, wherein/f only when the distance between the search point and the connection line of the feature points is less than or equal to the threshold value_r-f_j/. ltoreq.delta.the result is accepted, otherwise this characteristic point is considered notBelonging to this set.

(2) The shot sample retrieval is applied to a food safety tracing system, a video library containing 200 shots is established, the shots are taken from four links of food production, processing, circulation and consumption, recording is started from the links of food planting, cultivation and production processing, and the whole process from farmland to dining table tracking and tracing are realized, wherein the whole information comprises all information in the circulation processes of transportation, packaging, subpackaging, sale and the like. And (3) giving food shots needing to be inquired, and displaying the inquiry result according to the distance (as shown in figure 3). The result shows that the retrieval obtains better results and the retrieval precision is more accurate.

In conclusion, the food safety tracing system searching method based on lens sample searching is completed. The method has better time sequence, universality and relevance, improves the retrieval precision by extracting the key frame and combining the spatial distance through approximation, reduces the complexity and reduces the time required by retrieval.

Claims (4)

1. A food safety tracing system retrieval method based on lens sample retrieval is characterized in that: the method comprises the following steps:

A. acquiring a food shot picture to be retrieved, segmenting a shot to extract a key frame, and forming a feature space capable of fully reflecting the content of the shot according to a feature point set of the key frame;

B. connecting the characteristic points to form a curve, performing linear approximation on the curve, and determining the optimal approximation of the points of inflection as a key frame;

C. and determining the distances between the retrieval image and all the characteristic connecting lines in the sample lens through the threshold values of the characteristic point connecting lines, and sequencing the distances to obtain the most similar lens so as to finish the retrieval of the food safety tracing system.

2. The lens sample retrieval-based food safety tracing system retrieval method of claim 1, characterized in that: the step A comprises the following steps:

(1) acquiring food shot pictures to be searched, dividing shots, extracting key frames, finding out a boundary between two continuously appearing shots through detecting shot switching points, gathering the frames belonging to the same shot together, and gathering feature points in each frame to form a feature space capable of fully reflecting the content of the shot;

(2) two key frames F are selected to represent the main content of the shot_a、F_bCorresponding to two feature points f_a、f_b，Is provided with

Where d is the dimension of the feature space, the similarity between two frames is measured by the euclidean distance:

D(f_a,f_b)＝∥f_a-f_b∥。

3. the lens sample search-based food safety tracing system search method according to claim 1 or 2, characterized in that: the step B comprises the following steps:

(1) n feature points in the shot feature set S, structureA connecting line, if there are Q sets in the space, the total characteristic line isProjection of search point r on feature connecting line

f_r＝f_a+k(f_a-f_b)

Wherein,therefore, the similarity between the search point and the shot is:

D(f_r,(f_a,f_b))＝∥r-f_r∥

(2) performing linear approximation on the characteristic point connecting curve by using an approximation line l_al_bApproximation curve f_af_bThe feature space is composed of n-1 feature lines, the line of continuous reverse curve points is used for approaching the feature curve, and the frame corresponding to the reverse curve points is used as the key frame of the lens; setting a threshold value ε, if D (f)_j,f_j+1)>ε or D (f)_j-₁,f_j+1)>c is e, then f_j+1Is a new key frame, where 0<j is n-1, c is a constant and 1<c<2, searching key frames in each feature space, determining the number of the key frames in the shot, wherein the number of the key frames can reflect the features of the shot, and if no key frame exists, the whole shot can be regarded as a feature point; if the number of the key frames is large, the shot content is changed greatly or the set threshold is small, the threshold can be adjusted for recalculation, and if the feature points of the new key frame are on a straight line, the middle redundant feature points can be removed, so that the calculation complexity is reduced.

4. The lens sample retrieval-based food safety tracing system retrieval method of claim 3, wherein: the step C comprises the following steps:

(1) setting a threshold value delta of the distance between the connection lines of the feature points, wherein/f only when the distance between the search point and the connection line of the feature points is less than or equal to the threshold value_r-f_j/. ltoreq.δ, the result is accepted, otherwise the characteristic point is not considered to belong to the set;

(2) the method comprises the steps of searching and applying a shot sample to a food safety tracing system, establishing a video library containing a plurality of shots, recording all information including transportation, packaging, subpackaging and sales circulation processes from four links of food production, processing, circulation and consumption, giving food shots needing to be inquired, and displaying an inquiry result according to a distance.