CN206363347U - Based on Corner Detection and the medicine identifying system that matches - Google Patents

Based on Corner Detection and the medicine identifying system that matches Download PDF

Info

Publication number
CN206363347U
CN206363347U CN201720282609.1U CN201720282609U CN206363347U CN 206363347 U CN206363347 U CN 206363347U CN 201720282609 U CN201720282609 U CN 201720282609U CN 206363347 U CN206363347 U CN 206363347U
Authority
CN
China
Prior art keywords
medicine
microprocessor
image
memory
gray
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.)
Active
Application number
CN201720282609.1U
Other languages
Chinese (zh)
Inventor
金捷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiaxing Wenda Mdt Infotech Ltd
Original Assignee
Jiaxing Wenda Mdt Infotech Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Jiaxing Wenda Mdt Infotech Ltd filed Critical Jiaxing Wenda Mdt Infotech Ltd
Priority to CN201720282609.1U priority Critical patent/CN206363347U/en
Application granted granted Critical
Publication of CN206363347U publication Critical patent/CN206363347U/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Abstract

The utility model discloses a kind of based on Corner Detection and the medicine identifying system that matches.The microprocessor electrically connects the camera, memory and loudspeaker independently of each other simultaneously.The camera is used to gathering the image of medicine, and by the medicine image transmitting collected to the microprocessor.The microprocessor obtains medicine image, and is processed as gray-scale map, and the microprocessor detection medicine gray-scale map obtains several angle points and above-mentioned angle point is described.Complete to match two images by calculating the Hamming distance of two images, so as to identify corresponding medicine, there is higher recognition rate and accuracy rate under various complex environments, the crowd of dysopia can be preferably helped.

Description

Based on Corner Detection and the medicine identifying system that matches
Technical field
The utility model belongs to computer vision and machine learning techniques field, and in particular to one kind based on Corner Detection with The medicine identifying system of matching.
Background technology
According to incompletely statistics, about more than 300,004,200 ten thousand people are perplexed by vision disorder in the world.Depending on Power obstacle refers to visual impairment, disabled or defect, table caused by aging, eye illness and eye such as are wound at the reason congenital and day after tomorrow Now the composition or function for the organs of vision (including eyeball, ophthalmic nerve) and brain optic nerve maincenter occur some or all of Obstacle, so that visual impairment or the visual field diminish patient's eyes to some extent, it is impossible to which things is correctly recognized to external world.According to According to relevant regulations, dysopia patient eyesight is below 0.3 or the visual field is no more than 20 °.
Dysopia is broadly divided into blind person and the class of amblyopia patient two.Aided according to the defect of amblyopia patient and the vision used Utensil, amblyopia patient normally behaves as following a few classes:One is insensitive to solid, it is impossible to normally recognize three-dimensional body;Two be scarce The weary identification to light, changes insensitive to light;Three be insensitive to color, it is impossible to recognizes normal color, as colour blindness; Four be that patient's field range is too small, it is impossible to surrounding is experienced as normal person.Due to vision system in various degree it is impaired, Disabled or defect, amblyopia patient can use vision to be engaged in certain work, studying and living, but in different aspect, depending on Feel system has certain obstacle, it is necessary to complete normal study, work by certain vision auxiliary equipment and live.
Total blindness patient then can not use vision system to carry out normal working and learning completely, and they can only pass through human body Other sense organs such as tactile and the sense of hearing come replace vision carry out daily life.At present compared to total blindness crowd, China's dysopia Crowd is more, they can be carried out mostly by certain Information barrier-free technology and vision auxiliary equipment normal study and Life, but be due to that most vision auxiliary equipment function is single and on the high side, do not spread to the vision in city and rural area Obstacle crowd, service rate is relatively low.The vision auxiliary equipment of high efficiency low cost how is developed, the service of disturbance people auxiliary implement is improved Rate and quality of life are an important research directions of researcher, are also society and national urgent problem.
Corner Detection and extraction are the premises of graphical analysis and image recognition, are sign and the description of image local feature, Light and shade changes what lines in violent part, representative image intersected in its part that often lines intersect in correspondence image and image Part and the larger flex point of curvature.Angle point is and affected by environment due to its preferable antijamming capability and higher robustness It is smaller.In recent years, the detection method of angle point also achieves many great achievements in recent years.Adaptive universal Accelerated fractionation is examined It is relatively good Corner Detection to survey (Adaptive and Generic Accelerated Segment Test, AGAST) algorithm One of algorithm.The algorithm optimizes the greedy searching algorithm ID3 decision-makings used in FAST algorithms using efficient binary decision tree Tree, by merging two decision trees, according to present image information self-adapting using most efficiently decision tree, drastically increases The stability and detection rates of algorithm.FAST and AGAST algorithms rapidly can detect and extract image local Corner Feature, The real-time processing of frame of video is met, but is due to that only gray scale is handled, angle point does not have rotational invariance and Scale invariant Property.Quick retina key point (Fast Retina KeyPoint, FREAK) algorithm is the feature point description for mobile device Algorithm, also needs to improve in the case where the internal memory and processor resource of system are relatively fewer.
Utility model content
The utility model can not in usually living for dysopia crowd there is provided one kind for the situation of prior art Well identification medicine and design it is a kind of based on Corner Detection and the medicine identifying system that matches.
The utility model uses following technical scheme, described to be included based on Corner Detection with the medicine identifying system matched:
Camera, microprocessor, memory and loudspeaker, the microprocessor described in electrical connection while take the photograph independently of each other As head, memory and loudspeaker;
The camera is used to gathering the image of medicine, and by the medicine image transmitting collected to the microprocessor Device;
The microprocessor obtains medicine image, and is processed as gray-scale map, and the microprocessor detects medicine gray-scale map Obtain several angle points and above-mentioned angle point is described;
The memory built-in has drug data bank, and the microprocessor deposits Feature Descriptor and corresponding medicine information Above-mentioned drug data bank is stored in, the microprocessor calculates Hamming distance according to the Feature Descriptor of training image and target image To match medicine image, the microprocessor will recognize that obtained medicine information is exported by the speaker sound, to remind Dysopia crowd.
According to above-mentioned technical proposal, smart mobile phone is respectively adopted in the camera, microprocessor, memory and loudspeaker Camera, microprocessor, memory and loudspeaker.
The utility model is disclosed based on Corner Detection and the medicine identifying system that matches, and its advantage includes:For Existing AGAST algorithms are improved, and are improved algorithm and are become in dimensional variation, rotationally-varying, smear out effect, illumination variation and compression Robustness and stability in terms of change, cope with interference of different nature, improve Corner Detection speed.Meanwhile, simplify The sampling model of FREAK algorithms, makes it quickly and efficiently to be run on low internal memory and the mobile device calculated so that system A variety of medicines can be rapidly and accurately identified, there is higher recognition rate under various complex environments and accurate Rate, the crowd of dysopia can be helped well.
Brief description of the drawings
Fig. 1 is multiple dimensioned AGAST algorithm flows schematic diagram of the present utility model.
Fig. 2 is that FREAK of the present utility model describes sampling model in algorithm.
Fig. 3 A and Fig. 3 B are the sampling models of the FREAK algorithms after the utility model is improved.
Fig. 4 is the characteristic direction of the utility model FREAK arthmetic statements.
Fig. 5 is the system flow chart of the utility model preferred embodiment.
Fig. 6 A and Fig. 6 B are the medicine image angle point detects schematic diagrams of the utility model preferred embodiment.
Fig. 7 is the medicine images match schematic diagram of the utility model preferred embodiment.
Fig. 8 A, Fig. 8 B and Fig. 8 C are the medicine identification experimental result of the utility model preferred embodiment under various circumstances.
Embodiment
The utility model discloses a kind of based on Corner Detection and the medicine identifying system that matches, with reference to being preferable to carry out Example, is further described to embodiment of the present utility model.
Referring to Fig. 1 to Fig. 4 of accompanying drawing, recognized using described based on Corner Detection with the medicine of the medicine identifying system matched Method comprises the following steps:
Step one (medicine image Corner Detection):Medicine image is detected according to the multiple dimensioned AGAST algorithms of adaptive threshold Corner Feature, obtain several angle points;
Step 2 (medicine image corner description):Operator is described according to improved FREAK, binaryzation feature describing word is used The angle point of above-mentioned medicine image is described symbol string;
Step 3:Circulation carries out step one and step 2, to be trained to medicine image, sets up Feature Descriptor and medicine The corresponding drug storage of product information;
Step 4:According to training image and the Feature Descriptor of target image, the Hamming distance for calculating two images has been come Into matching two images, so as to identify corresponding medicine.
Preferably, camera collection image is used in the step one, the gray-scale map of image is marked, used AGAST algorithms obtain angle point to detect;
Preferably, extracted in the step one to improve the robustness and stability of detection algorithm in multiscale space The feature of image, and adaptive gray threshold is provided with according to the contrast of image.
Further, comprise the following steps in the step one:
Step 1.1:Angle point mark and the adaptive gray threshold t of detection are calculated according to the contrast of image;
In AGAST algorithms, when carrying out rough detection to angle point on gray level image, the detection threshold that gray level image is marked Value t is general unrelated with image information, is typically manually set.In detection algorithm, gray threshold t representatives are detected between angle point Minimal gray difference, also illustrates that the maximum gradation value that detection algorithm can ignore that to the noise in image, threshold value t setting can shadow The number of the angle point detected is rung, excessive or very few angle point can all influence the training of decision tree, finally reduce detection efficiency.It is right In the image of different contrast, different efficiency are often shown using identical threshold value.The high image of contrast, is detected Angle point number is more, may be descriptive bad, and the low image of contrast, the angle point number detected is few, influences the training of decision tree, Grey value profile herein according to image is improved to gray threshold, the preceding n maximum that is sorted according to image intensity value and Last n minimum value meets the gray threshold of image information to calculate, and is adaptively adjusted according to the contrast of different images Gray threshold, makes the angle point detected neither influence the training of decision tree, while having stronger descriptive, formula (1) is described The computational methods of image adaptive gray threshold:
Wherein, n represents the number for the pixel chosen, and α represents weight, typically takes 0.05 to 0.2.IjmaxAnd IjminRespectively Represent the preceding n pixel after gray value descending sort and rear n pixel.
Step 1.2:Light and shade mark is carried out to image according to the adaptive gray threshold of image, by gray level image labeled as bright Light and shade changes violent point as angle point candidate point in dark different region, image, and Binary decision is set up by angle point candidate point Set to detect other angle points;Formula (2) describes the computational methods of image light and shade mark:
Step 1.3:The multiscale space of image is built, the angle point of detection image in multiple dimensioned description sequence;
The multiscale space of image, i.e. gaussian pyramid are built by the smoothing processing of image.In order to reduce algorithm Complexity, image represented using two groups of Gaussian convolution images, in this two groups of images are divided into according to the sampling scale of primary image Tomographic image ciWith middle tomographic image di.Original image down-sampled obtains intermediate layer most bottom as the bottom of internal layer by its 1.5 times Layer, then to different tomographic images are organized using different yardstick progress Gaussian Blurs, with the ratio between the adjacent two layers image of group The example factor is 2, thus constitutes some continuous images of row yardstick, the gaussian pyramid of image is built with this, is then used AGAST Corner Detection Algorithms extract the feature of Scale invariant on pyramid.
Further, referring to Fig. 1 of accompanying drawing, comprise the following steps in step 1.3:
Step 1.3.1:In the adjacent two layers and the adjacent two layers of different groups of images with group image, non-maxima suppression is used Candidate point is screened, the angle point of the regional stability is selected, to suppress multiple angle points occur in the same area;
Step 1.3.2:According to the corner location on different scale, the optimum angle of angle point is gone out using least square fitting Degree, and the yardstick of fitting is corrected using one-dimensional parabola, calculate the yardstick and positional information for obtaining characteristic point.
According to above preferred embodiment, the step 2 is specially:The sampled point of neighborhood is chosen according to the position of characteristic point Carry out the description character string of construction feature point, in order to ensure rotational invariance, characteristic point is calculated by the half-tone information of characteristic point Gradient direction, the feature point description character string built and the information of medicine are finally completed both by nomenclature of drug it Between mapping.
Further, comprise the following steps in the step 2:
Step 2.1:Feature neighborhood of a point is sampled, characteristic point is described selected pixels point;
Referring to Fig. 2 of accompanying drawing, the sampling configuration of FREAK algorithms imitates the human eye retina region knot in human visual system Structure, sampled point is generated with this.This mode is more nearly the mode that human eye retina obtains image, and sampled point is evenly distributed on In concentric circles centered on characteristic point, the distribution with photosensory cell in human eye retina's structure is similar.Distance sample is nearer Region, sampling point density is bigger, and the radius of Gaussian Blur is smaller, is mainly used in obtaining high-resolution image, and extract figure The minutia of picture;The more remote region of distance feature point, the density of sampled point is smaller, and the radius of Gaussian Blur is bigger, main to use In the relatively low image of acquisition resolution ratio, and extract image general frame and edge contour information.Have in Fig. 2 it is many it is of different sizes simultaneously There is overlapping circle, represent different sample areas, the center of circle is the sampled point in the region, and radius represents the standard deviation of Gaussian Blur, By the way that more information can be obtained using different Gaussian blurring functions to overlapped region, so that characteristic point is retouched It is more accurate and unique to state.
Fig. 3 A and Fig. 3 B, the FREAK operator of accompanying drawing are participated in generally using 8 layers of retina sampling structure, from characteristic point by inner Outside sampling point distributions are:1,6,6,6,6,6,6,6, wherein 1 is characterized a little, other centers of circle are sampled point, using characteristic point as On the concentric circles in the center of circle, 6 sampled points are uniformly distributed, and the radius of Gaussian kernel is directly proportional to the radius for the circle that sampled point is the center of circle. In order to obtain the more features of image, the receptive field of neighbouring sample point (i.e. centered on sampled point, Gaussian kernel is the circle of radius) It is designed as overlapped structure.According to contour features such as angle point, the edges that target is primarily upon in object matching identification process, The missing of the minutias such as the texture of target influences little to the result of matching, in order to improve the speed of service and save in operation Deposit, according to retina image-forming principle, 8 layers of sampling structure are reduced to 5 layers by the utility model preferred embodiment, are formed after improving FREAK operator is described, its characteristic point is distributed as:1,6,6,6,6.
Step 2.2:By thresholding build binaryzation FREAK character string is described, by receptive field to Gaussian Blur after Difference comparsion cascade is formed;
Vacation lets f be FREAK binary systems and describes character string, then according to formula (3):
Wherein, PaIt is receptive field pair, N is the length that desired binary system describes character string, T (Pa) expression formula see formula (4):
WhereinThe gray value that sampled point is obtained after Gaussian Blur is represented respectively.r1,r2Respectively Represent radius of the receptive field to front and rear receptive field.
Step 2.3:According to the information of feature vertex neighborhood gray scale, feature neighborhood of a point gradient is calculated;
Referring to Fig. 4 of accompanying drawing, in order to ensure that description has rotational invariance, FREAK operators use similar BRISK operators Mode, be that each characteristic point builds principal direction information.Due to there is 25 sampled points around characteristic point, can produce 25* (25-1)/ 2=300 sampled point pair, chooses wherein 33 long, symmetrical sampled point to extracting the direction of characteristic point, characteristic point herein Neighborhood gradient calculation mode is shown in formula (5):
Wherein, the gradient information of O features vertex neighborhood, M represents the quantity of sampled point pair, G represent groups of samples into set, PoRepresent the position of each pair sampled point.
Further, the step 3 and step 4 are specially:According to the binaryzation between target image and training image Feature Descriptor, description string by 0 and 1 coding constitute, so carry out images match, compare description character string when, by by Position xor operation (XOR) come calculate two description character string Hamming distances (Hamming Distance), when two character strings Hamming distance when being less than the similar threshold value of two images, then two images match successes, otherwise match unsuccessful, Hamming distance Calculation see formula (6):
Referring to Fig. 5 of accompanying drawing, the training part with the medicine identifying system matched based on Corner Detection with medicine With the identification division of medicine, wherein:
The training part of the medicine:The angle point of medicine image is entered first by the multiple dimensioned AGAST algorithms after improvement Row detection, then carries out feature description to the angle point detected using the FREAK algorithms after optimization, builds the description of binaryzation feature Son, finally sets up the medicine information drug data bank corresponding with Feature Descriptor.
The identification division of the medicine:Target image first in detection video, using the above method to the figure in video Detection and description as carrying out angle point, are then matched description in description and drug data bank of generation, to know Do not go out medicine, describe the mapping relations between son and medicine information finally according to medicine, medicine information is extracted from database, lead to The user that the medicine information that voice will identify that passes to dysopia is crossed, helps them preferably to live.
According to above preferred embodiment, it is described camera is included based on Corner Detection and the medicine identifying system that matches, it is micro- Processor, memory and loudspeaker, the microprocessor electrically connect the camera, memory and raised one's voice independently of each other simultaneously Device;
The camera is used to gathering the image of medicine, and by the medicine image transmitting collected to the microprocessor Device;
The microprocessor obtains medicine image, and is processed as gray-scale map, and the microprocessor is according to adaptive threshold Multiple dimensioned AGAST algorithms detection medicine gray-scale map obtain several angle points, describing operator according to improved FREAK uses two-value Change feature describes character string and above-mentioned angle point is described;
The memory built-in has drug data bank, and the microprocessor deposits Feature Descriptor and corresponding medicine information Above-mentioned drug data bank is stored in, the microprocessor calculates Hamming distance according to the Feature Descriptor of training image and target image To match medicine image, the microprocessor will recognize that obtained medicine information is exported by the speaker sound, to remind Dysopia crowd.
Preferably, it is above-mentioned can be by means of mobile device, such as with China based on Corner Detection and the medicine identifying system matched For the smart mobile phone that Ascend P7 are representative, the camera, microprocessor, memory of smart mobile phone common configuration is respectively adopted And loudspeaker.By taking common drug Granmaoling Granules as an example, using Huawei Ascend P7 as medicine identifying system carrier, mainly Parameter is:Operating system:Android OS 4.4, cpu frequency 1.8GHz, the pixel of rear camera 13,000,000, experimental threshold values are 20, tested by training different number of drug storage accuracy and the testing time of medicine, it is as a result as shown in the table:
According to above preferred embodiment, compared with prior art, requirement of the utility model to equipment is relatively simple, operation It is more convenient, handle, all improved a lot in speed and discrimination for interference of different nature.Referring to accompanying drawing Fig. 6 A, Fig. 6 B, Fig. 7 and Fig. 8 A, Fig. 8 B, Fig. 8 C, can be seen that Corner Detection and matching algorithm after improving from experimental result All there is stronger robustness and stability, medicine to picture rotation, partial occlusion and the weaker situation of illumination in two systems The speed of product identification within the specific limits, is influenceed smaller, the recognition rate and accuracy of medicine are equal by the number of picture to be identified Meet and require, the purpose of disturbance people vision auxiliary can be realized.
For a person skilled in the art, the technical scheme described in foregoing embodiments can still be repaiied Change, or equivalent substitution is carried out to which part technical characteristic, all within spirit of the present utility model and principle, that is made is any Modification, equivalent substitution, improvement etc., should be included in protection domain of the present utility model.

Claims (2)

1. it is a kind of based on Corner Detection and the medicine identifying system that matches, it is characterised in that including:
Camera, microprocessor, memory and loudspeaker, the microprocessor electrically connect the shooting independently of each other simultaneously Head, memory and loudspeaker;
The camera is used to gathering the image of medicine, and by the medicine image transmitting collected to the microprocessor;
The microprocessor obtains medicine image, and is processed as gray-scale map, and the microprocessor detection medicine gray-scale map is obtained Several angle points and above-mentioned angle point is described;
The memory built-in has drug data bank, and Feature Descriptor and corresponding medicine information are stored in by the microprocessor Above-mentioned drug data bank, the microprocessor according to the Feature Descriptor of training image and target image calculate Hamming distance with Power dispensers image, the microprocessor will recognize that obtained medicine information is exported by the speaker sound, to remind vision Obstacle crowd.
2. it is according to claim 1 based on Corner Detection and the medicine identifying system that matches, it is characterised in that the shooting Camera, microprocessor, memory and the loudspeaker of smart mobile phone is respectively adopted in head, microprocessor, memory and loudspeaker.
CN201720282609.1U 2017-03-22 2017-03-22 Based on Corner Detection and the medicine identifying system that matches Active CN206363347U (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201720282609.1U CN206363347U (en) 2017-03-22 2017-03-22 Based on Corner Detection and the medicine identifying system that matches

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201720282609.1U CN206363347U (en) 2017-03-22 2017-03-22 Based on Corner Detection and the medicine identifying system that matches

Publications (1)

Publication Number Publication Date
CN206363347U true CN206363347U (en) 2017-07-28

Family

ID=59380611

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201720282609.1U Active CN206363347U (en) 2017-03-22 2017-03-22 Based on Corner Detection and the medicine identifying system that matches

Country Status (1)

Country Link
CN (1) CN206363347U (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109800632A (en) * 2018-12-11 2019-05-24 广东省结核病控制中心 A kind of drug administration behavior recognition methods and system for tuberculosis patient

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109800632A (en) * 2018-12-11 2019-05-24 广东省结核病控制中心 A kind of drug administration behavior recognition methods and system for tuberculosis patient

Similar Documents

Publication Publication Date Title
CN106980852B (en) Based on Corner Detection and the medicine identifying system matched and its recognition methods
CN103632136B (en) Human-eye positioning method and device
CN104050471B (en) Natural scene character detection method and system
CN103902961B (en) Face recognition method and device
CN103902958A (en) Method for face recognition
CN105740780B (en) Method and device for detecting living human face
CN106709450A (en) Recognition method and system for fingerprint images
CN102844766A (en) Human eyes images based multi-feature fusion identification method
CN108229458A (en) A kind of intelligent flame recognition methods based on motion detection and multi-feature extraction
CN106778664A (en) The dividing method and its device of iris region in a kind of iris image
CN106056064A (en) Face recognition method and face recognition device
CN103218615B (en) Face judgment method
CN108363957A (en) Road traffic sign detection based on cascade network and recognition methods
CN103839056B (en) A kind of method for recognizing human eye state and device
CN109815850A (en) Iris segmentation and localization method, system, device based on deep learning
CN106874825A (en) The training method of Face datection, detection method and device
CN106096527A (en) A kind of recognition methods of real-time high-precision online bank note face amount
CN109035227A (en) The system that lung tumors detection and diagnosis is carried out to CT image
CN206363347U (en) Based on Corner Detection and the medicine identifying system that matches
CN104050455B (en) A kind of skin color detection method and system
Manaf et al. Color recognition system with augmented reality concept and finger interaction: Case study for color blind aid system
Zabihi et al. Vessel extraction of conjunctival images using LBPs and ANFIS
Niu et al. Automatic localization of optic disc based on deep learning in fundus images
CN111401145A (en) Visible light iris recognition method based on deep learning and DS evidence theory
CN106960199A (en) A kind of RGB eye is as the complete extraction method in figure white of the eye region

Legal Events

Date Code Title Description
GR01 Patent grant
GR01 Patent grant