WO2022005274A1

WO2022005274A1 - A system and method for detecting a moving object in a dynamic environment from a sequence of captured images

Info

Publication number: WO2022005274A1
Application number: PCT/MY2020/050185
Authority: WO
Inventors: Kim Meng Liang; Hock Woon Hon; Durai Naidu A/L RANGGANAZAN
Original assignee: Mimos Bhd
Current assignee: Mimos Bhd
Priority date: 2020-06-30
Filing date: 2020-12-01
Publication date: 2022-01-06
Anticipated expiration: 2022-12-30
Also published as: MY205788A

Abstract

The present invention discloses a system to detect a moving object in a dynamic environment from a sequence of captured images, comprising a learning rate module (9), an image filtering module (10), an object likelihood level module (11), and an object identification module (12). The system executes the following steps: determining a learning rate seed (301) based on the speed rate (306) of the moving object, applying the determined learning rate seed (301) to a data training module (13) for filtering of background images (403) from captured images, calculating an object likelihood level (505) by computing statistical properties of the filtered images (406), and identifying the moving object based on the object likelihood level (505), wherein the learning rate seed (301) and the object likelihood level (505) are updated based on a processing of one or more new images.

Description

A SYSTEM AND METHOD FOR DETECTING A MOVING OBJECT IN A DYNAMIC ENVIRONMENT FROM A SEQUENCE OF CAPTURED IMAGES

FIELD OF INVENTION

The present invention relates to a system and method for detecting a moving object in a dynamic environment, and more particularly to a system and method for detecting the moving object from a sequence of captured images.

BACKGROUND OF THE INVENTION

In recent years, advancements in technology has brought forth the integration of artificial intelligence in digital industry. Image processing technologies combined with machine learning algorithms have drastically improved the information extraction capabilities of a surveillance system. In the surveillance system, cameras, video cameras or other sensing devices are commonly used to capture images or videos of the environments. Conventionally, the captured images or videos are then analysed by an operator to understand and monitor activities surrounding the environment. As such, automated surveillance systems are developed to reduce the human error that occurs during analysis by the operator, by combining the image processing technology and machine learning algorithms in artificial intelligence, thus further improving quality of detection and processing.

Many technologies related to detecting a moving object in a dynamic environment have been proposed to further improve the system. For example, United States with patent publication no. US8159536B2 discloses a method for detecting desired objects in a highly dynamic environment by a monitoring system, comprising: taking at least one image of the environment; generating at least one background image from the image taken, wherein the background image comprises a plurality of pixels; dividing the background image into a plurality of pixel blocks; forming at least one data cluster of each pixel block based on at least one feature of the background pixels in the pixel block; comparing each pixel in the pixel block to a subsequently generated background image to generate a first and second discrepancy value; determining whether the pixel of the subsequent image is a background pixel or foreground pixel based on the first and second discrepancy value; generating a binary map of the subsequent image based on the subsequent image based on the determined foreground pixels and background pixels; and marking connected pixels in the binary map corresponding to the foreground pixels to form at least one foreground object, wherein the foreground object is at least one object detected in the environment.

A United States patent with publication no. US20190005361A1, recites a method of real-time identification moving objects in video images, the method comprises the following steps: capturing a video input from a scene comprising one or more moving objects, wherein the video input comprises at least two temporally spaced image frames captured from the scene; transforming the video input into one or more image pattern layers, wherein each of the image pattern layers comprises a pattern representing one of the desired moving objects; determining a probability match between each of the image pattern layers and a stored image in a big data library; adding one or more image pattern layers having the probability of match that exceeds a predetermined level to the big data library automatically; and outputting the probability of match. The method of obtaining a result through probability matching is unreliable as the outcome may not be accurate to the object of interest.

Another technology as disclosed in United States patent with publication no. US7929730B2 recites a method and system for object detection and tracking, for detecting and tracking a plurality of continuous videos generated by the object motion in a scene, wherein the method comprising of: updating information of a foreground and background appearance models based on a new input video and past accumulated information; performing an object detection process based on the information of the updated foreground and background appearance models, past propagated time axis information and current observed video; using the detection result from the object detection process to update the information of the foreground and background appearance models until a convergent condition is met; using the convergent detection result as a new tracking measurement to update the tracking prediction model; using the updated tracking prediction model to obtain a final tracking result; and predicting the location of the tracked object at the next scene, storing the predicted labelling and appearance information of the foreground and background objects to update further.

The aforementioned patent documents disclose the various systems and methods of detecting and tracking an object based on performing probability matches and predictions from a plurality of continuous images. However, these systems and method fail to provide an effective detection on the moving object of interest in a highly dynamic environment.

Accordingly, it would be desirable to provide a system and method for detecting a moving object in a dynamic environment, more particularly a system and method employing machine learning algorithms based on a predetermined feature of the moving object to effectively identify the moving object in the foreground.

SUMMARY OF INVENTION

The present invention discloses a system to detect a moving object in a dynamic environment from a sequence of captured images, comprising of: a learning rate module, configured to determine a learning rate seed based on the speed rate of the moving object; an image filtering module, configured to apply the determined learning rate seed to a data training module for filtering of background images from captured images; an object likelihood module, configured to calculate a likelihood level by computing statistical properties of the filtered images; an object identification module, configured to identify the moving object based on the likelihood level; and wherein the learning rate seed and the object likelihood level are updated based on a processing of one or more new images.

Preferably, the image filtering module applies an absolute subtraction technique to remove the background from the captured images.

Preferably, the object likelihood level module is further configured to compute statistical properties of the filtered images based on an average calculation of a series of indicator values for each filtered image.

Preferably, the data training module is configured to optimise the updated learning seed as a new learning rate seed to be input into the learning rate module.

In another aspect of this invention, there is provided a method for detecting a moving object in a dynamic environment from a sequence of captured images, the method comprising: determining a learning rate seed based on a speed rate of the moving object, by a learning rate module; applying the determined learning rate seed to a data training module for filtering of background images from the captured images, by an image filtering module; computing a likelihood level by computing statistical properties of the filtered images, by an object likelihood level module; and identifying the moving object based on the likelihood level, by an object identification module; and wherein the learning rate seed and the object likelihood level are updated based on processing of one or more new images.

Preferably, the method further comprising applying, by the image filtering module, an absolute subtraction technique to filter the background images from the captured images. Preferably, the absolute subtraction technique comprises the steps of applying two or more updated background images onto the captured images; and removing the updated background images from the overall captured images.

Preferably, the method further comprising the step of computing, by the object likelihood level module, statistical properties of the filtered images based on an average calculation of a series of indicator values for each filtered image.

Preferably, the series of indicator values for each filtered image is calculated by the steps of determining a standard deviation value of the filtered images; computing an occupancy value for pixels of interest based on an overlay of a dynamic grid base on the filtered images; and calculating a scatteredness value for the pixels of interest based on the distance of the pixels of interest on the dynamic grid base to adjacent points on the dynamic grid base.

One skilled in the art will readily appreciate that the present invention is well adapted to carry out the objects and obtain the ends and advantages mentioned, as well as those inherent therein. The embodiment described herein is not intended as limitations on the scope of the invention.

BRIEF DESCRIPTION OF THE DRAWINGS

For the purpose of facilitating an understanding of the invention, there is illustrated in the accompanying drawing the preferred embodiments from an inspection of which when considered in connection with the following description, the invention, its construction and operation and many of its advantages would be readily understood and appreciated. FIG. 1 is a block diagram of a system to detect a moving object in a dynamic environment from a sequence of captured images, according to the present invention.

FIG. 2 is a flow chart illustrating an exemplary embodiment for a method to detect a moving object in a dynamic environment from a sequence of captured images based on the above-mentioned system.

FIG. 3 is a flow chart illustrating a preferred embodiment to determine the learning rate seeds based on the speed rates of the moving object.

FIG. 4 is a flow chart illustrating a preferred embodiment to filter the background images based on the determined learning rate seeds.

FIG. 5 is a flow chart illustrating an exemplary embodiment to calculate the object likelihood level and the varying filtered images corresponding to the object likelihood levels.

FIG. 6 is a flow chart illustrating a preferred embodiment to identify the moving object based on the object likelihood levels.

FIG. 7 is an exemplary embodiment for determining the optimum learning rate from each learning rate seed based on the object likelihood level.

FIG. 8 is a flow chart illustrating an exemplary embodiment for determining the optimum learning rate for each learning rate seed by conducting forward and backward learning rate updates.

DETAILED DESCRIPTION OF THE INVENTION It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, that execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer- readable memory produce an article of manufacture including instruction means that implement the fimction/act specified in the flowchart and/or block diagram block or blocks.

The computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that execute on the computer or other programmable apparatus provide steps for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. The invention will now be described in greater detail, by way of example, with reference to the drawings.

Referring to FIG. 1, there is provided a system for detecting a moving object in a dynamic environment from a sequence of captured images from an image capturing unit 1, wherein the image capturing unit 1 may be a digital or analog camera, digital video recorder, a network video camera or any other mediums suitable for capturing images in a sequence. In one preferred embodiment, the captured images are transferred over the internet 2, wherein the captured images may be transferred wirelessly over a communication network to a recording unit 3. The communication network can be a wireless network connection established via a wireless protocol cloud such as Long- Term Evolution (LTE) cloud, Code Division Multiple Access (CDMA) and its derivatives, Enhanced Data Rates for GSM Evolution (EDGE), 3G protocol, High Speed Packet Access (HSPA), 4G protocol, 5G protocol and the like, in accordance to the advancement of wireless technology with time. The communication network can also be a wired network via a local network port. Further, the recording unit 3 may be a video converter which operates by converting the data from the image capturing unit 1 to be stored in a mass storage unit 4, wherein the mass storage unit 4 may be an internal memory storage device which can be installed within the image capturing unit or externally linked to the image capturing unit 1. The mass storage unit can be any type of hardware device that is capable of storing and retrieving data, wherein the hardware device may come in the form of hard drives or optical discs. Primary storage devices, such as Raw Access Memory (RAM), is used by computer systems to temporarily store and retrieve data whereas secondary storage devices, such as Hard Disk Drives (HDD) or Solid State Drives (SSD) store data permanently. Further, the mass storage unit may also be a third party cloud service provider, wherein the third party cloud service provider is able to transmit and store data to remote storage system, where it is maintained, managed, backed up and made available to users over a network, typically the internet 2.. Further, the captured images transferred over to the internet 2 are subsequently transferred into a video acquisition unit 5, wherein the video acquisition unit further processes the captured images into a sequence of images to be configured in a configuration unit 6 and object detection unit 8. Preferably, the configuration unit 6 configures at least one data analytics of either the image capturing unit 1 or the video acquisition unit 5. For example, the data analytics to be configured on the image capturing unit 1 may comprise of camera credentials, location name of the camera, or stream URLs, whereas the data analytics to be configured on the captured images from the image capturing unit 1 may be a monitoring area which includes a region of interest, or a data analytics parameter such as image resolution or pixel density. Preferably, the configured data from the configuration unit 6 is transferred to a database 7, which then transfers said information to the object detection unit 8, wherein the database 7 may be an internal or external computer system in wired or wireless communication with the video acquisition unit 5, configuration unit 6 and object detection unit 8. Preferably, the plurality of modules in the object detection unit 8 communicate with one another to form an interconnected communication network.

In one particular embodiment, the object detection unit 8 comprises a learning rate module 9, configured to determine a learning rate seed 301 based on the speed rate 306 of the moving object. Preferably, the speed rate 306 of the moving object may be classified into three groups, which is low 306a, medium 306b, and high 306c, whereby a corresponding learning rate to be initiated for the respective groups. Further, the learning rate seed 301 may be adjusted based on the optimum learning rate seed 303 from the previous frame. As such, the learning rate module 9 will determine the initial learning rate seeds 301 for each speed rate 306 defined and input the various learning rate seeds 301 into the next module.

In another particular embodiment, the object detection unit 8 comprises an image filtering module 10, configured to filter the background from the background images 403 that are updated based on the predefined learning rate seeds 301. With reference to A in FIG. 4, each learning rate seed 301 determined by the various speed rates 306 of the moving objects subsequently updates the background images 403 which are then compared to the current image frame of the captured images. Preferably, the image filtering module 10 applies an absolute subtraction 404 technique to filter and remove the background images 403 from the current image frame, which is subsequently compared to a predefined threshold 405 implemented into the image filtering module 10 to obtain a final filtered image 406 for the corresponding learning rate seed 301.

In another particular embodiment, the object detection unit 8 comprises an object likelihood module 11, configured to compare differences between the filtered images 406 based on an average calculation of a series of indicator values for each filtered image 406. Referring to FIG. 5, the object likelihood module 11 determines the standard deviation 501 of the filtered image 406 formed as the first indicator value. The second indicator value is computed by the occupancy 502 of pixels of interest 507 within a dynamic grid base 509 when overlaid onto the filtered image 406. The third indicator value is obtained by calculating the scatteredness 503 of the pixels of interest 507 within the dynamic grid base 509 based on the distance of each pixel of interest 507 to adjacent points on the dynamic grid base 509. All three indicator values are then weighted and averaged to produce a likelihood level of varying degrees depending on the filtered image 406 formed, which will be discussed further herein. By way of example, variants of the filtered images 406 corresponding to the respective object likelihood levels 505 are shown in FIG. 5.

In another particular embodiment, the object detection unit 8 comprises an object identification module 12 for identifying the object of interest based on the calculated object likelihood levels 505. The varying filtered images 406 are compared based on their likelihood levels in order to determine the final output image of the moving object. By way of example, a filtered image 406 with a high 510c likelihood level corresponding to one learning rate seed 301 is compared to other filtered images 406 of high 510c likelihood levels corresponding to other learning rate seeds 301. Upon comparison, the filtered images 406 with high 510c likelihood levels are combined to determine the final image outcome which successfully depicts the moving object in a foreground image. FIG. 2 illustrates an exemplary embodiment for a method for detecting a moving object in a dynamic environment from a sequence of captured images based on the above- mentioned system. At Step 201, the image capturing unit 1 first captures a sequence of images to be used for detecting the moving object. At Step 202, the learning rate module 9 determines the learning rate seed 301 based on the speed rate 306 of the moving object, wherein the speed rate 306 of the moving object is predefined based on prior knowledge of the environment. However, if prior knowledge of the environment does not exist, then three common speed rates 306 are used: low 306a, medium 306b, high 306c. At Step 203, the image filtering module 10 applies the learning rate seed 301 to the data training module 13 in order to determine a suitable learning rate at incremental 402 interval and decremental 401 interval, and subsequently begin the process of filtering the background images 403 from the sequence of captured images. At Step 204, the filtered images 406 are then obtained from the image filtering module 10 based on a comparison with a predefined threshold 405. At Step 205, the object likelihood level module 11 calculates the object likelihood level by computing the statistical properties for each filtered image 406. At Step 206, the object likelihood levels 505 obtained from the filtered images 406 are compared to a threshold at Step 206 to determine if another set of filtered images 406 is required. Steps 207 and 208 are taken if the object likelihood levels from the filtered images 406 are categorised as low 701 likelihood level. At Step 211, the object identification module 12 then identifies the moving object based on the calculated likelihood level 505. The calculated object likelihood level 505 of the moving object is then subsequently applied to the data training module 13 to update the learning rate seed 301, wherein the updated learning rate seed 301 for the identified moving object is optimised 303 as a new learning rate seed 301, and wherein the steps taken to obtain a filtered image 406 is repeated. Step 209 is taken if there are no optimum learning rate seeds 303 that could be obtained from the initial learning rate seeds 301, whereby a forwards 802 and backwards 801 update on the learning rate seeds 301 will take place until an optimum learning rate seed 303 is achieved. At Step 212, all filtered images 406 from each optimum learning rate seed 303 will be obtained and subsequently combined at Step 213 into a master filtered image 214, wherein the master filtered image 214 establishes the successfully identified moving object. The steps shown in FIG. 2 are further elaborated in FIG. 3 to FIG. 8.

FIG. 3 illustrates an exemplary embodiment to determine the learning rate seeds 301 for detecting the moving object. At step 300, a minimum learning rate value 304 and maximum learning rate value 305 are generated by the learning rate module 9. The learning rate seed 301 is initially determined by three common speed rate 306 groups of the moving object at different predefined locations in the first frame, whereby the speed rates 306 are classified as low 306a, medium 306b and high 306c speed rates 306. Learning rate seeds 301a, 301b, 301c are then determined based on the respective speed rates 306 for the initial frame. For a following frame in Step 302, if a previous optimum learning rate seed 303 is within the tolerance range of the predefined learning rate seed 301 at the specific object speed rate 306, then no new learning rate seed 301 is generated and the following frame will utilize the optimum learning rate seed 303 to become a reference learning rate seed 301 for the next module. However, the initial predefined learning rate seeds 30 Id, 30 le will be generated and initialized if the optimum learning rate seed 303 is not within the tolerance range of the predefined learning rate seeds 301 from the previous frame.

FIG. 4 illustrates an exemplary embodiment to filter the background images 403 by removing the background from the captured images. At step 400, each learning rate seed 301 further generates two learning rates at incremental 402a, 402b, 402c intervals and decremental 401a, 401b, 402c intervals from the learning rate seed 301. In one preferred embodiment, the incremental 402a, 402b, 402c learning rates and decremental 401a, 401b, 401c learning rates have a predefined range from the initial learning rate seeds 301. At step 403, the background image is updated based on each learning rate intervals 401, 402. With the updated background images 403 formed, the image filtering module 10 filters the background images 403 from the current image by applying an absolute subtraction 404 to obtain a filtered image 406 which only shows the foreground object. The output images of the absolute subtraction 404 technique are then compared to a threshold at step 405 in order to obtain the final filtered image 406 corresponding to the learning rate seed 301.

FIG. 5 illustrates an exemplary embodiment to calculate the likelihood level 501 of the moving object. The object likelihood level module 11 computes three indicator values to be averaged in order to obtain a likelihood level 505 of the object of interest. At step 501, the standard deviation of the obtained filtered image 406 is determined, giving a first indicator value. At step 502, the second indicator value is computed by overlaying a dynamic grid base 509 onto the filtered image 406 in order to calculate the occupancy of pixels of interest 507 on the dynamic grid base 509, wherein each pixel of interest 507 indicates a segment of the moving object which is located on the dynamic grid base 509. Step 508 shows by way of example, the occupancy value of the pixels of interest 507 of the moving object in an 8x8 dynamic grid base 509, wherein the pixels of interest 507 corresponding to the moving object in the filtered image 406 occupies 7/64 pixels. At Step 503, the third indicator value is calculated based on the scatteredness 506 of the pixels of interest 507 of the filtered image 406 in the dynamic grid base 509. By way of example but not limitation, the scatteredness 506 value is obtained by calculating the distance of the pixels of interest 507 to adjacent points in the dynamic grid base 509, whereby an average is calculated from the average distance of the pixels of interest 507 with the total number of calculated distances from the adjacent points. The first two steps of calculating the distance of the pixels of interest 507 to the adjacent points in the dynamic grid base 509 and calculating the average of the average distances are repeated with all points of interest. As such, the final scatteredness 506 value is obtained by averaging all the average values from the first three steps. At step 504, the final object likelihood level 505 is obtained by calculating the weightage and average of all the indicator values. Each object likelihood level 505 obtained is further classified in order of lower likelihood level to higher likelihood level, as depicted in 510 which shows the variants of filtered images 406 corresponding to the object likelihood levels 505. The variant filtered images 406 classified in respective categories of low 510a likelihood level, medium 510b likelihood level, and high 510c likelihood level are exemplified further in FIG. 6 and FIG. 7.

FIG. 6 illustrates an exemplary embodiment for obtaining an absolute subtracted 404 image if two filtered images are of non-high likelihood levels. Step 601 depicts a mixture of low 510a likelihood level and medium 510b likelihood level examples of movement mixtures of the different objects in the actual scenario undergoing the interval learning rate process to generate a new filtered image 406 which will be input into the object likelihood module 9 to determine the respective likelihood level 505. The absolute subtraction 404 technique will not be conducted if there is a pairing of varying levels, whereby a filtered image 406 pairing which does not contain any high 510c likelihood level in any one of them will not undergo the absolute subtraction 404 technique. However, should a pairing of filtered images 406 exhibit the same object likelihood levels 505, either low 510a likelihood level or medium 510b likelihood level without any high 510c likelihood level, the object likelihood module 11 will apply the absolute subtraction 404 technique to obtain a possible filtered image 406 of high 510c likelihood level, which will be used to determine the moving object.

FIG. 7 illustrates an exemplary embodiment of determining the optimum learning rate seed 303 from a sequence of difference images 406 of varying object likelihood levels 505, based on each learning rate seed 301. The three categories of the object likelihood levels 505 as shown are low 510a likelihood level, medium 510b likelihood level, and high 510c likelihood level. The filtered images 406 are categorised as such that the filtered images 406 are of low 510a likelihood level if only noise is detected, medium 510b likelihood level if both noise and the moving object are detected in the filtered images 406, and high 510c likelihood level if the filtered images 406 contain only the moving object. At Step 701, one of the exemplary results of all the filtered images 406 of each learning rate seed 301 is all the likelihood level is medium 510b and another exemplary result is there is a mixture of low 510a and medium 510b likelihood levels. These two examples indicate that the optimum learning rate has yet to be achieved. However, at Step 702, the optimum learning rate seed 303 is achieved if at least one of the filtered images 406 are of a high 510c likelihood level or if all the filtered images 406 have a low 510c likelihood level.

FIG. 8 illustrates an exemplary embodiment of determining the optimum learning rate seed 303 by implementing backward 803 and forward 804 updates on the learning rate seeds 301, as configured in the data training module 13. At Step 801, in the event that no optimum learning rate seed 303 is achieved as all filtered images 406 from the respective learning rate seeds 301 corresponds to low 510a likelihood level, the data training module 13 will update the learning rate seeds 301 in a manner of forward 804a, 804b, 804c updates or backward 803a, 803b, 803c updates in order to obtain a new set of learning rate seeds 301, wherein at least one new pair of learning rates are similarly created from the learning rate seeds 301 in order to obtain a new sequence of filtered images 406. At Step 802, the process of backward 803 updates and forward 804 updates on the learning rate seed 301 will take place until at least one filtered image 406 with a high 510c likelihood level is obtained, indicating the achievement of the optimum learning rate seed 303. Once the optimum learning rate seed 303 is achieved, it will be input into the learning rate module 1 to be utilized for the following sequence of captured images.

The present disclosure includes as contained in the appended claims, as well as that of the foregoing description. Although this invention has been described in its preferred form with a degree of particularly, it is understood that the present disclosure of the preferred form has been made only by way of example and that numerous changes in the details of construction and the combination and arrangements of parts may be resorted to without departing from the scope of the invention.

Claims

1. A system for detecting a moving object in a dynamic environment from a sequence of captured images, the system is characterized by having: a learning rate module (9), configured to determine a learning rate seed (301) based on a speed rate (306) of the moving object; an image filtering module (10), configured to apply the determined learning rate seed (301) to a data training module (13) for filtering of background images (403) from the captured images; an object likelihood level module (11), configured to calculate a likelihood level (505) by computing statistical properties of each filtered image (406); and an object identification module (12), configured to identify the moving object based on the object likelihood level (505), wherein the learning rate seed (301) and the object likelihood level (505) are updated based on a processing of one or more new images.

2. The system according to Claim 1, wherein the image filtering module (10) is further configured to remove the background from the captured images by applying an absolute subtraction (404) technique.

3. The system according to Claim 1, wherein the object likelihood level module (11) is further configured to compute statistical properties for each filtered image (406) based on an average calculation of a series of indicator values for each filtered image (406).

4. The system according to Claim 1, wherein the data training module (13) is configured to optimise the updated learning seed (301) as a new learning rate seed (301) to be input into the learning rate module (9).

5. A method for detecting a moving object in a dynamic environment from a sequence of captured images, the method is characterized by having the steps of: determining, by a learning rate module (9), a learning rate seed (301) based on a speed rate (306) of the moving object; applying, by an image filtering module (10), the determined learning rate seed (301) to a data training module (13) for filtering of background images (403) from the captured images; computing, by an object likelihood level module (11), a likelihood level (505) by computing statistical properties of the filtered images (406); and identifying, by an object identification module (12), the moving object based on the object likelihood level (505), wherein the learning rate seed (301) and the object likelihood level (505) are updated based on a processing of one or more new images.

6. The method according to Claim 5, further comprising the step of applying, by the image filtering module (10), an absolute subtraction technique (404) to filter the background images (403) from the captured images.

7. The method according to Claim 6, wherein the absolute subtraction technique (404) comprises the steps of: applying two or more updated background images (403) onto the captured images; and removing the updated background images (403) from the captured images.

8. The method according to Claim 5, further comprising the step of computing, by the object likelihood level module (11), statistical properties of the filtered images (406) based on an average calculation a series of indicator values for each filtered image (406).

9. The method according to Claim 8, wherein the series of indicator values for each filtered image (406) is calculated by the steps of: determining a standard deviation (501) value of the filtered images (406); computing an occupancy (502) value for pixels of interest (507) based on an overlay of a dynamic grid base (509) on the filtered images (406); and calculating a scatteredness (506) value for the pixels of interest (507) based on the distance of the pixels of interest (507) on the dynamic grid base (509) to adjacent points on the dynamic grid base (509).

10. The method according to Claim 5, further comprising the step of updating the learning rate seed (301) in predefined increment values (804) and decrement values (802) until an optimum learning rate seed (303) is achieved to be input into the learning rate module (9) as a new learning rate seed (301).