WO2012060682A1

WO2012060682A1 - Method of self-enrollment for object identification

Info

Publication number: WO2012060682A1
Application number: PCT/MY2011/000077
Authority: WO
Inventors: Kim Meng Liang
Original assignee: Mimos Bhd
Current assignee: Mimos Bhd
Priority date: 2010-11-02
Filing date: 2011-06-06
Publication date: 2012-05-10
Anticipated expiration: 2013-05-02
Also published as: MY177015A

Abstract

A method of performing self-enrollment for object identification, the methods comprising the steps of computing (11) an enrollment model which includes the steps of detecting (14) all the object for a predetermined duration, extracting (15) visual features, determining authorized objects, storing (19) them in feature space, and normalising (24) feature space for obtaining optimum enrollment model, and providing feedback (12) to said enrollment model on the detected unauthorized object and updating (12) said enrollment model for any new or removal of authorized object during the object identification in the system.

Description

Method of Self-Enrollment: for Object Identification

Field of Invention The present invention relates generally to an object identification, more particularly to a method for performing self-enrollment for object identification.

Background of the Invention

Various of high security systems for object identification are highly needed in securing buildings, especially the buildings contain with high density of authorized and unauthorized objects. Most of the systems are using features identification methods, which are the methods of measuring and digitizing information which are unique to the individual or object, such as colour, textures, fingerprints, facial recognition, iris recognition or palm prints. After scanned, the information is converted into a digital code which is stored in a central database on which the system runs and where the information can be retrieved for verification and matching.

Object identification with manual enrollment process is not effective for unstable movement and various types of authorized object. Unstable movement involves frequently exists new unauthorized object where previous is authorized. Furthermore, authorized object has many types, including the permanent and temporary. With these complexities of movement and type of objects in the building, object identification task is a highly challenging task if manual enrollment process is needed.

As disclosed in the US Patent No. 20040170323 and PCT application No. WO2008155313, manual enrollment process is needed prior to object identification. Therefore, the solutions from these prior arts are irrelevant for providing object identification solutions for buildings especially with high density of authorized and unauthorized objects.

It is an objective of the present invention to provide an object identification system with self-enrollment process to assist the security personnel to carry out the video monitoring task.

Other objects of this invention will become apparent on the reading of this entire disclosure.

Summary of the Invention

In the present invention, a method for performing self- enrollment for object identification purpose -is described. The method comprising processes of extracting visual features, storing them in feature space and normalizing feature space to obtain optimum enrollment model. Once the enrollment model is computed, real time object identification is carried out. In object identification process, the enrollment model obtained feedback from the users of the system on the detected unauthorized object. The enrollment model is continuously updated for any new or removal of authorized object in the system.

Brief Description of the Drawings

Other objects, features, and advantages of the invention will be apparent from the following description when read with reference to the accompanying drawings. In the drawings, wherein like reference numerals denote corresponding parts throughout the several views:

Figure 1 shows the overview of the object identification process of the present invention;

Figure 2 illustrates a flowchart of the method of computing enrollment model of the present invention;

Figure 3 is a flowchart showing the method of normalizing feature space; Figure 4 depicts a profile for duration and frequency for each location in the feature space;

Figure 5 shows a process of calculating the feature value for each profile; igure 6 illustrates the profile generation and optimum alue calculation; and

Figure 7 is a flowchart which illustrates the method of performing feedback and update to the enrollment model.

De-tailed Description of the Preferred Embodiments

In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be understood by those of ordinary skill in the art that the invention may be practiced without these specific details. In other instances, well-known methods, procedures and/or components have not been described in detail so as not to obscure the invention. Reference will now be made in detail to the preferred embodiments of the present invention, examples of which are illustrated in the accompanying drawings. The present invention is directed to a method for performing self-enrollment for object identification system (10) . The present method (10) provides a technique to compute enrollment model (11) automatically at predetermined duration prior to the object identification process, therefore valid objects are not needed to enroll themselves and the object identification is done base on the generated enrollment model as shown in Figure 1. Visual features such as colour, texture and shape are extracted for each detected object at a predetermined duration and stored in a multi-dimensional feature space. The feature space is normalized for obtaining optimum enrollment model which represents the information of the authorized objects. After the enrollment model is computed (11), the system may perform real time object identification. During the object identification process, the enrollment model will obtain feedback (12) from the users of the system on the detected unauthorized object and continuously update for any new or removal of authorized object in the system. The method of computing the enrollment model is shown in Figure 2. .In the beginning, the present cycle number and duration are extracted (13) . The present cycle number represents the running cycle number as compared to the maximum cycle number threshold (T2) that equivalent to the predetermined duration for computing the enrollment model. Whereas the present cycle duration represents the running cycle duration as compared to the maximum cycle duration threshold (Tl) that equivalent to the total duration in one cycle number.

After that, objects which appeared in the scene will be detected (14) and their features will be extracted (15). The object detection will be performed base on the known detection techniques such as for example the motion detection and foreground detection. Then the duration of the object appeared in the scene will be extracted (16). With the extracted features, the location of the object will be computed (17) in the multi-dimensional feature space. Each location has duration and frequency as the parameters. For each calculated location, the duration and frequency of the appeared object are accumulated (18).

If the present cycle duration is lesser than the Tl value, the steps of extracting the present cycle number and cycle duration (13) until the step of accumulating the duration and frequency of the appeared object (18) are repeated. Otherwise, the accumulated duration and frequency (18) are stored (19) in a database. Then the present cycle number is validated with T2 value (20) . If the present cycle number is lesser than T2 value (21), the present cycle number is increased and the cycle duration is initialized (22). The steps of extracting the present cycle number and cycle duration (13) is repeated again until the validation of present cycle number with T2 value (20) . If the present cycle number is larger than T2 value (23), the feature space is normalised (24) and the steps are shown in Figure 3.

The steps of normalising feature space begins with initializing a T3 value (25) where T3 value is the distance threshold that represents the maximum distance between two locations in feature space to be integrated. This is then followed by the step of integrating (26) pair of locations in the feature space into new integrated location if the distance between the locations is lesser than T3 value. The duration and frequency which is correspondent to the pair of locations are added into the new integrated location (27) .

Then the feature space with the new integrated locations and added parameters of duration and frequency are stored (28) in the database. For each stored feature space, the profiles of duration and frequency, respectively for each location are generated (29) which is shown in Figure 4. In Figure 4, the values of duration and frequency for each cycle of the total maximum cycle number are generated into graph representation, respectively. The profiles are generated for each location in the feature space. The right illustration annotated as A in Figure 4 depicts three examples of profile. After that, it follows by the step of calculating (30) the feature value for each generated profile for the duration and frequency as shown in Figure 5. In Figure 5, two thresholds are used namely a high threshold (HT) and a low threshold (LT) . The profile that contained between these two threshold values is considered in the feature calculation as highlighted on the right illustration B in Figure 5. The calculated feature includes area under the curve that is darkened. For each feature space, each location is validated base on the calculated features from the profiles of duration and frequency, in order to reflect the significant of the location as authorized object for object identification purpose.

If both of the features are larger than the predetermined thresholds as annotated as T4, the number of location threshold, the location is counted (31) as valid location and the total valid location are stored as number of location. If number of location is larger than the minimum number of location as annotated as T5 (32), the minimum location threshold, the T3 value is increased (33) . Then the step of integrating the locations (26) until the step of validating the calculated number of location (31) is repeated.

Otherwise, the profile of the number of location with feature value larger than T4 for all the T3 values is generated (34) . With the profile, the optimum T3 value is calculated (35) as shown in Figure 6. The optimum T3 value is determined base on the global highest value of number of location and the feature space that relates to the optimum T3 value is chosen as the enrollment model for object identification. The locations in this feature space represent the authorized objects for object identification purpose. In the chosen feature space, the live span and validity for each location are initialize (36) with a predetermined maximum values. Then the chosen feature space is stored (37) as enrollment model.

Figure 7 shows a flowchart which illustrates the method of performing feedback and update (12) to the enrollment model. In the beginning, the status for all the locations in the enrollment model is initialized to False (41) . Then the cycle duration is extracted (42) and the objects in the scene are detected (43) . The features of the detected objects are extracted (44) and with the extracted features, the location in the feature space known as enrollment model is calculated (45). The methods used for extracting the cycle duration (42), detecting objects (43), extracting features (44) and calculating location in feature space (45) are similar to the methods used in the steps of computing enrollment model.

If the calculated location is similar to the available locations in the enrollment model (46) , the validity is compared (47) with the predefined maximum value. If the validity is equivalent (48) to the predefined maximum value, the status of the calculated location is assigned to True (49) . Otherwise, if the calculated location is not similar to the available locations in the enrollment model (50), the detected object is highly possible is an unauthorized object. Then the feedback on the unauthorized object is obtained (51) .

If the feedback is positive (52) where the unauthorized object is a valid object, the calculated location is created (53) as new location in the enrollment model. For this new location, the live span and validity are initialized (54) to maximum or minimum value. After the calculated location is validated and updated in the enrollment model, the initial extracted cycle duration is compared (55) with Tl value, which is equivalent to the total duration in one cycle number as similar to the threshold used in computing enrollment model.

If the cycle duration is lesser than Tl value (56) , the step of extracting cycle duration (42) until the step of comparing cycle duration with Tl value (55) is repeated. Otherwise (57), the status of each location in the enrollment model is validated (58) . If the status is not equivalent to True (59), the life span of the location is subtracted (60) by one. Then the location is removed (61) from the enrollment model if the life span is equivalent to zero (62) . With the update and feedback during object identification process, the enrollment is continuously updated to reflect uthorized object throughout time and at the same able to detect unauthorized object. As will be readily apparent to those skilled in the art, the present invention may easily be produced in other specific forms without departing from its essential characteristics. The present embodiments is, therefore, to be considered as merely illustrative and not restrictive, the scope of the invention being indicated by the claims rather than the foregoing description, and all changes which come within therefore intended to be embraced therein.

Claims

1. A method of performing self-enrollment for object identification, the methods comprising the steps of:

computing (11) an enrollment model which includes the steps of:

detecting (14) all the object for a predetermined duration; extracting (15) visual features;

determining authorized objects, storing (19) them in feature space; and

normalising (24) feature space for obtaining optimum enrollment model; and

providing feedback (12) to said enrollment model on the detected unauthorized object and updating (12) said enrollment model for any new or removal of authorized object during the object identification in the system.

2. The method of performing self-enrollment for object identification as claimed in claim 1, wherein said step of determining authorized objects further comprising the steps of extracting (16) the duration of the object appeared in the scene, calculating (17) the location of said object and accumulating (18) the duration and frequency of the appeared object.

3. The method of performing self-enrollment for object identification as claimed in claim 1, wherein said step of normalising (24) feature space includes the steps of grouping detected objects based on difference distances between locations of detected objects in said feature space and verifying said grouped objects as authorized objects based on the features of the grouped objects.

4. The method of performing self-enrollment for object identification as claimed in claim 3, wherein said step of grouping detected objects includes the process of integrating (26) locations with distance between locations less than the distance threshold; combining (27) duration and frequency for each integrated location; and storing (28) them in a database.

5. The method of performing self-enrollment for object identification as claimed in claim 4, wherein said step of grouping detected objects further comprising the process of generating (29) profile for all the cycles in locations; and calculating (30) the feature values for each profile.

6. The method of performing self-enrollment for object identification as claimed in claim 3, wherein said step of verifying said grouped objects includes the process of determining (31) each location in each feature space as valid location based on the location threshold and the total valid location is stored as number of location.

7. The method of performing self-enrollment for object identification as claimed in claim 6, wherein said process further comprising the steps of calculating (35) optimum distance threshold in the chosen feature space as the enrollment model.

8. The method of performing self-enrollment for object identification as claimed in claim 1, wherein said step of providing feedback (12) to said enrollment model includes the steps of extracting (42) the cycle duration; detecting (43) objects; extracting (44) features of the detected objects; and calculating (45) the location in feature space for determining unauthorized detected · obj ect .

9. The method of performing self-enrollment for object identification as claimed in claim 1, wherein said step of updating (12) said enrollment model includes the steps of validating and creating new calculated location of valid unauthorized object and removing of location from feature space if said location is failed for validation.