A method and system for implementing ASCII art visual privacy protection and fall detection

Abstract

The present invention discloses a method and system for implementing ASCII art visual privacy protection and fall detection, which relates to the technical fields of computer vision and pattern recognition. The method comprises the following steps: receiving an original fall behavior dataset, supplementing the original fall behavior dataset with Vincent video technology to obtain a supplemented fall behavior dataset, preprocessing the supplemented fall behavior dataset to obtain visual latent video data in which video frame images produce a visual privacy protection effect; extracting image features from the supplemented fall behavior dataset, inputting the image features into a pre-established support vector regression model, and outputting an image privacy protection score; supplementing lost features in the visual latent video data in which video frame images produce a visual privacy protection effect to obtain processed visual latent video data, inputting the processed visual latent video data into a pre-established YOLO+ posture estimation model, and outputting a fall behavior detection result.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and pattern recognition, and in particular to a method and system for implementing ASCII art visual privacy protection and fall detection. Background Art

[0002] The lack of reliable fall detection systems has long been a serious problem in home health monitoring. While video surveillance systems can monitor elderly people at home to detect falls, traditional approaches to this type of intelligent detection fail to address privacy concerns. Furthermore, many people prefer not to have their actions observed. Therefore, developing a fall detection algorithm that balances safety monitoring with privacy protection is of great practical significance and application value.

[0003] In recent years, research on fall detection has been gaining traction. Researchers both domestically and internationally have conducted in-depth studies on human behavior recognition and fall detection technologies. Based on different implementation methods, fall detection technologies can be broadly categorized into three types: Wearable device-based fall detection typically involves embedding sensors in the user's wearable device. While wearable devices are sensitive to their environment, prolonged wear can cause a degree of discomfort. Usability is challenging, as elderly individuals may forget to wear these devices in their daily lives. Furthermore, these sensors are prone to false alarms, making them unreliable. Scene-aware fall detection utilizes non-video sensors placed around pedestrian areas to capture human features and record elderly individuals' behavioral data. However, this approach has limited applicability due to its high cost, susceptibility to environmental interference such as noise, and high detection error rates. Furthermore, it is difficult to maintain, resulting in a high maintenance error rate. Computer vision-based fall detection utilizes fixed image or video capture devices at the detection site to collect image or video data. These devices are then processed to identify human behavior and determine whether a fall has occurred. This approach offers the advantages of simple installation, low cost, minimal disruption, and the ability to simultaneously record a variety of daily activities. However, traditional vision-based fall detection methods may infringe on the privacy interests of participants.

[0004] Currently, with the rapid development of Internet of Things (IoT) technology, cameras are widely deployed in many public spaces for surveillance purposes. However, the application of these methods in private homes has been questioned due to their potential privacy violations. Summary of the Invention

[0005] In order to address the deficiencies mentioned in the above background technology, the purpose of the present invention is to provide a method and system for implementing ASCII art visual privacy protection and fall detection.

[0006] In a first aspect, the purpose of the present invention can be achieved by the following technical solution: a method for implementing ASCII art visual privacy protection and fall detection, the method comprising the following steps:

[0007] Receiving an original fall behavior dataset, supplementing the original fall behavior dataset using Vincent video technology to obtain a supplemented fall behavior dataset, and preprocessing the supplemented fall behavior dataset to obtain visually hidden video data with a visual privacy protection effect on video frame images;

[0008] Extract image features from the supplementary fall behavior dataset, input the image features into a pre-established support vector regression model, and output the image privacy protection score;

[0009] The lost features in the visual and latent video data that produce the visual privacy protection effect of the video frame image are supplemented to obtain the processed visual and latent video data, which is input into the pre-established YOLO+ posture estimation model to output the fall behavior detection result.

[0010] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: the process of supplementing the original fall behavior dataset using Vincent video technology: adding multiple scenarios on the basis of the original fall behavior dataset, and also taking into account factors such as the age, gender, clothing, scene location, etc. of the elderly at home, and using a scene recognition model to verify the feasibility of the supplemented dataset, and finally obtaining a supplemented fall behavior dataset.

[0011] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: the process of preprocessing the supplementary fall behavior dataset includes: dividing the video data into positive and negative sample sets according to whether a fall behavior occurs in the video clip, and marking corresponding category labels on the positive and negative sample sets respectively, and using a scene recognition model to evaluate the practicality of the supplementary dataset video data, and then performing ASCII art encoding processing on the expanded video data based on ASCII art technology, and finally obtaining visually hidden video data in which the video frame image produces a visual privacy protection effect.

[0012] In conjunction with the first aspect, in certain implementations of the first aspect, the method further includes: a calculation process of performing ASCII art encoding processing on the extended video data based on the ASCII art technology is as follows:

[0013] First, grayscale is performed. The color of each pixel in the color image is determined by three components: Red, Green, and Blue. Each component has 255 possible median values. When these three components are the same, the desired grayscale image X is obtained. Then, the grayscale image X of M×N pixels is divided into n identical m×n small blocks according to the proportion of the selected font; that is, X=[x1,x2,...x n ], each such block will be output as a character, for each pixel block, select a character, the average value of the pixel block of its gray value Y i A good approximation of , and then a block average operation is performed on each pixel block as follows:

[0014]

[0015] where Z i As the sum of the i-th pixel block of image X, is the spatial size of the i-th pixel of image X, Y i is the average value of the i-th pixel block in the grayscale image X.

[0016] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: extracting image features in the supplementary fall behavior dataset by extracting significant structural features SSF and dense color features DCF.

[0017] In conjunction with the first aspect, in certain implementations of the first aspect, the method further includes: a process of supplementing the lost features in the visually hidden video data that produces the visual privacy protection effect on the video frame image:

[0018] The SYM-Track framework is used to accurately and effectively segment and track the main objects in the visual and latent video data with greatly reduced image quality, and to extract the spatiotemporal characteristics of the motion trajectory of the main objects in the visual and latent video frames.

[0019] In conjunction with the first aspect, in certain implementations of the first aspect, the method further includes: inputting the processed visual and hidden state video data into a pre-established YOLO+-based posture estimation model to output a fall behavior detection result:

[0020] YOLOv7 is used for target detection, and the detected target box information is then input into the Pose model for pose estimation to obtain accurate target detection and pose estimation results. The YOLOv7+ pose estimation model used is based on a deep convolutional neural network architecture. It accepts input images and generates feature maps, which are then used to predict the key point positions of each person. After tracking the target in the ASCII art-encoded fall video, fall detection is performed on the processed visual and latent video data to obtain fall behavior detection results.

[0021] In a second aspect, in order to achieve the above-mentioned objectives, the present invention discloses an ASCII art visual privacy protection and fall detection implementation system, comprising:

[0022] a data processing module, configured to receive an original fall behavior dataset, supplement the original fall behavior dataset using Vincent video technology to obtain a supplemented fall behavior dataset, and preprocess the supplemented fall behavior dataset to obtain visually hidden video data that produces a visual privacy protection effect for video frame images;

[0023] The privacy protection module is used to extract image features from the supplementary fall behavior dataset, input the image features into a pre-established support vector regression model, and output an image privacy protection score;

[0024] The fall detection module is used to supplement the lost features in the visual and latent video data that produce the visual privacy protection effect of the video frame image, obtain the processed visual and latent video data, input the processed visual and latent video data into the pre-established YOLO+ posture estimation model, and output the fall behavior detection result.

[0025] In conjunction with the second aspect, in certain implementations of the second aspect, the system further includes: a process in the data processing module of supplementing the original fall behavior dataset using Vincent video technology: adding multiple scenarios to the original fall behavior dataset, taking into account factors such as the age, gender, clothing, and scene location of the elderly living at home, and using a scene recognition model to verify the feasibility of the supplemented dataset, thereby ultimately obtaining a supplemented fall behavior dataset;

[0026] Alternatively, the process of preprocessing the supplementary fall behavior dataset in the data processing module includes: dividing the video data into positive and negative sample sets according to whether a fall behavior occurs in the video clip, marking corresponding category labels on the positive and negative sample sets respectively, and using a scene recognition model to evaluate the practicality of the video data of the supplementary dataset, and then performing ASCII art encoding processing on the expanded video data based on ASCII art technology, and finally obtaining visually hidden video data with a visual privacy protection effect for the video frame image;

[0027] Alternatively, the calculation process of performing ASCII art encoding processing on the expanded video data based on ASCII art technology in the data processing module is as follows:

[0028] First, grayscale is performed. The color of each pixel in the color image is determined by three components: Red, Green, and Blue. Each component has 255 possible median values. When these three components are the same, the desired grayscale image X is obtained. Then, the grayscale image X of M×N pixels is divided into n identical m×n small blocks according to the proportion of the selected font; that is, X=[x1,x2,...x n ], each such block will be output as a character, for each pixel block, select a character, the average value of the pixel block of its gray value Y i A good approximation of , and then a block average operation is performed on each pixel block as follows:

[0029]

[0030] where Z i As the sum of the i-th pixel block of image X, is the spatial size of the i-th pixel of image X, Y i is the average value of the i-th pixel block of the grayscale image X;

[0031] Preferably, the privacy protection module extracts image features from the supplementary fall behavior dataset by extracting significant structural features SSF and dense color features DCF;

[0032] Preferably, the process of supplementing the lost features in the hidden state video data that produces the visual privacy protection effect for the video frame image in the fall detection module is as follows:

[0033] The SYM-Track framework is used to accurately and effectively segment and track the main objects in the visual and latent video data with greatly reduced image quality, and to extract the spatiotemporal characteristics of the motion trajectory of the main objects in the visual and latent video frames;

[0034] Preferably, the fall detection module inputs the processed visual and hidden video data into a pre-established YOLO+ posture estimation model to output the fall behavior detection result:

[0035] YOLOv7 is used for target detection, and the detected target box information is then input into the Pose model for pose estimation to obtain accurate target detection and pose estimation results. The YOLOv7+ pose estimation model used is based on a deep convolutional neural network architecture. It accepts input images and generates feature maps, which are then used to predict the key point positions of each person. After tracking the target in the ASCII art-encoded fall video, fall detection is performed on the processed visual and latent video data to obtain fall behavior detection results.

[0036] In another aspect of the present invention, in order to achieve the above-mentioned purpose, a terminal device is disclosed, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The memory stores a computer program capable of running on the processor, and when the processor loads and executes the computer program, it adopts the above-mentioned ASCII art visual privacy protection and fall detection implementation method.

[0037] Beneficial effects of the present invention:

[0038] The present invention introduces ASCII art technology into the traditional vision-based fall detection system and encodes the monitoring video. While eliminating visual information to produce a visual privacy protection effect, it also adds a variety of scenes and character features to the traditional elderly fall detection dataset, enriching the diversity of the dataset. This not only solves the problem of privacy infringement in the current vision-based monitoring method, but also improves the real-time performance of fall detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0040] Figure 1 It is a schematic flow chart of the method of the present invention;

[0041] Figure 2 Schematic diagram of the video frame effect of the supplementary data set in an embodiment of the present invention;

[0042] Figure 3 Schematic diagram of the visual privacy effect of an image based on ASCII art encoding in an embodiment of the present invention;

[0043] Figure 4 Schematic diagram of the visualization result of the motion trajectory of the target subject in the visually hidden video frame according to an embodiment of the present invention;

[0044] Figure 5 Schematic diagram of visualization of fall detection using hidden video in an embodiment of the present invention;

[0045] Figure 6 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0047] Example 1:

[0048] The following is an introduction to the relevant terms involved in the embodiments of this application:

[0049] ASCII Art: This primarily computer-generated art form uses computer characters (primarily ASCII) to represent images. It can be generated by a text editor. Many ASCII art artworks require a balanced font (a fixed-width font, such as that used on traditional typewriters) for display. ASCII art is used in situations where text is more stable and faster to display than images. This includes typewriters, teletypes, non-graphics terminals, early computer networks, email, and Usenet news messages. The simplest ASCII art consists of two or three letters, using text to express emotions.

[0050] like Figure 1 As shown, a method for implementing ASCII art visual privacy protection and fall detection includes the following steps:

[0051] S101: receiving an original fall behavior dataset, supplementing the original fall behavior dataset using Vincent video technology to obtain a supplemented fall behavior dataset, and preprocessing the supplemented fall behavior dataset to obtain visually hidden video data that produces a visual privacy protection effect for video frame images;

[0052] The process of using Vincent video technology to supplement the original fall behavior dataset: a variety of scenarios were added to the original fall behavior dataset, and factors such as the age, gender, clothing, and scene location of the elderly at home were also taken into consideration. The scene recognition model was used to verify the feasibility of the supplemented dataset, and finally a supplemented fall behavior dataset was obtained.

[0053] Using the Wensheng video technology of the diffusion model, we edited the public fall behavior video data based on the description of the required output, and made up for the lack of scenes and quantity in the traditional dataset. We added a variety of scenes (such as kitchen, bedroom, bathroom) and character features to the traditional elderly fall detection dataset. Figure 2The following example shows the generated video frames. Based on whether a fall occurs in the video clip, the set of videos containing falls is defined as the positive sample set, while the set of videos containing only normal daily behavior without falls is defined as the negative sample set.

[0054] The scene recognition approach involves transfer learning using the InceptionV3 architecture with ImageNet pre-trained weights, training this scene recognition network 200 times. We then used a nearly identical architecture, using Xception as the CNN architecture. Experiments revealed that the latter achieved significantly better accuracy. We then performed scene recognition on the generated video dataset to determine the feasibility of scene generation. We then compared the scene recognition accuracy of a subsequent ASCII art character set-encoded video dataset with the fall detection results. The fall detection performance was superior to the scene recognition performance, further demonstrating the privacy-preserving properties of ASCII art character set-encoded videos and the effectiveness of the encoding.

[0055] The preprocessing process of the supplementary fall behavior dataset includes: dividing the video data into positive and negative sample sets according to whether a fall behavior occurs in the video clip, marking the corresponding category labels on the positive and negative sample sets respectively, and using the scene recognition model to evaluate the practicality of the supplementary dataset video data. Then, based on the ASCII art technology, the expanded video data is ASCII art encoded, and finally the visually hidden video data with the visual privacy protection effect of the video frame image is obtained.

[0056] The calculation process of ASCII art encoding of the expanded video data based on ASCII art technology is as follows:

[0057] First, grayscale is performed. The color of each pixel in the color image is determined by three components: Red, Green, and Blue. Each component has 255 possible median values. When these three components are the same, the desired grayscale image X is obtained. Then, the grayscale image X of M×N pixels is divided into n identical m×n small blocks according to the proportion of the selected font; that is, X=[x1,x2,...x n ], each such block will be output as a character, for each pixel block, select a character, the average value of the pixel block of its gray value Y i A good approximation of , and then a block average operation is performed on each pixel block as follows:

[0058]

[0059] where Z i As the sum of the i-th pixel block of image X, is the spatial size of the i-th pixel of image X, Y i is the average value of the i-th pixel block of the grayscale image X. By using a limited set of different combinations of characters, it can depict different artistic effects and has low requirements for computing resources. Figure 3 As shown, the image quality after ASCII art encoding has hidden most of the details.

[0060] S102: Extracting image features from the supplementary fall behavior dataset, inputting the image features into a pre-established support vector regression model, and outputting an image privacy protection score;

[0061] A supplementary video dataset is established based on a public fall behavior dataset. Salient structural features (SSF) and dense color features (DCF) are extracted, and the subjective visual privacy protection labels are combined to train a support vector regression (SVR) model to evaluate the visual privacy protection of images.

[0062] S103: Supplement the lost features in the visual and latent video data that produce the visual privacy protection effect of the video frame image to obtain processed visual and latent video data, input the processed visual and latent video data into a pre-established YOLO+ posture estimation model, and output the fall behavior detection result.

[0063] The process of supplementing the missing features in the hidden video data to produce visual privacy protection effects on the video frame image:

[0064] The SYM-Track framework is used to accurately and effectively segment and track the main objects in the visual and latent video data with greatly reduced image quality, and to extract the spatiotemporal characteristics of the motion trajectory of the main objects in the visual and latent video frames.

[0065] The adopted "Segment and Track Anything" (SAM-Track) framework is a unified video segmentation framework that adopts a multimodal interactive approach and supports the combination of various tracking methods. It can selectively utilize interactive and automatic (unsupervised) tracking modes simultaneously.

[0066] SAM's interactive method (click or box selection) segments the object of interest in the reference frame. DeAOT then uses the segmentation results as annotations and uses a gated propagation module to hierarchically propagate visual embeddings and ID embeddings from past frames to the current frame for frame-by-frame object tracking. Although SAM is a powerful base model for interactive object segmentation, it also has certain limitations. Specifically, it does not provide enough semantic information, and text prompts may not be able to effectively support tasks that require a more detailed understanding of object segmentation.

[0067] To track new objects appearing in videos, two methods, SE (Segment Everything) and OIS (Object of Interest Segmentation), are proposed to obtain annotations for new objects in every n-th frame. In the SE method, we use the segment-everything function of SAM to obtain an object mask for each object in the key reference frame. DeAOT then tracks the new objects and the original objects based on the combined annotations. In contrast, the OIS method utilizes Grounding-DINO and SAM to obtain annotations for new objects appearing in videos. Specifically, Grounding-DINO detects objects, i.e., "person," in every n-th frame based on a predetermined textual cue. SAM and DeAOT then handle the annotation and tracking of new objects. We use Comparative Mask Results (CMR) to identify new objects. In CMR, we compare the tracking results of DeAOT with the annotation results of SAM in each key frame and select objects not tracked by DeAOT from the SAM annotations. Specifically, define N∈RH×W,T0∈RH×W,S∈RH×W and t as the new object mask, the background of DeAOT tracking results, the SAM annotation results and the minimum threshold of the new object, respectively, where H and W represent the height and width dimensions. The new object mask is obtained by using Eq. 2.

[0068] N=T0*S (2)

[0069] For an object x that appears in N, we define its size in S and N as x s and x n , if x s and x n If the ratio between is greater than t, the object will be defined as a new object, as expressed by Equation 3. By using CMR, we can significantly alleviate the problem that newly detected objects affect the ID tracking of tracked objects.

[0070]

[0071] SAM-Track, which combines interactive and automatic tracking modes, is what we ultimately need. The interactive tracking mode obtains annotations for the first frame of the video, while the automatic tracking mode tracks new objects that appear in the video and are not selected in the first frame. With the efficient DeAOT tracking mode, SAM-Track can quickly infer and track multiple objects. For videos that have undergone visual privacy protection, after SAM-Track target tracking, the movement trajectory of the target object in the video can be quickly discovered, which greatly reduces the feature loss problem caused by ASCII art encoding. It not only effectively protects privacy, but also can detect the behavior of the target object. Figure 4Shown are examples of tracked video frames.

[0072] The process of inputting the processed visual and hidden video data into the pre-established YOLO+ posture estimation model and outputting the fall behavior detection results:

[0073] YOLOv7 is used for target detection, and the detected target box information is then input into the Pose model for pose estimation to obtain accurate target detection and pose estimation results. The YOLOv7+ pose estimation model used is based on a deep convolutional neural network architecture. It accepts input images and generates feature maps, which are then used to predict the key point positions of each person. After tracking the target in the ASCII art-encoded fall video, fall detection is performed on the processed visual and latent video data to obtain fall behavior detection results.

[0074] Specifically, the present invention will be further described below through examples:

[0075] A binary statistical model between visual privacy protection and fall detection rate is established to balance the relationship between the two.

[0076] In order to balance the visual privacy protection and fall detection of ASCII art coding, this embodiment establishes a binary statistical model between the two. While achieving visual privacy protection of video data, it is still hoped to ensure the usefulness of the video after privacy protection, that is, to ensure its recognition effect in intelligent applications. When the visual privacy protection score of the image video frame is low, the visual information contained in the image video frame does not meet the privacy protection requirements. When the visual privacy protection score of the image video frame is high, the recognition rate of the image video frame will be greatly reduced, causing the image in the visual privacy protection state to lose its application value. Therefore, it is hoped to find a balance between the visual privacy protection score and the recognition rate, that is, to ensure a high recognition rate while maintaining high visual privacy protection, and to establish a quantitative correlation statistical model between visual safety and fall detection rate.

[0077] Example 2: The second aspect, as Figure 6 As shown, in order to achieve the above-mentioned purpose, the present invention discloses an ASCII art visual privacy protection and fall detection implementation system, comprising:

[0078] The data processing module 11 is configured to receive an original fall behavior dataset, supplement the original fall behavior dataset using Vincent video technology to obtain a supplemented fall behavior dataset, and preprocess the supplemented fall behavior dataset to obtain visually hidden video data that produces a visual privacy protection effect for video frame images.

[0079] The privacy protection module 12 is used to extract image features from the supplementary fall behavior dataset, input the image features into a pre-established support vector regression model, and output an image privacy protection score;

[0080] The fall detection module 13 is used to supplement the lost features in the visual and latent video data that produce the visual privacy protection effect of the video frame image, obtain the processed visual and latent video data, input the processed visual and latent video data into the pre-established YOLO+ posture estimation model, and output the fall behavior detection result.

[0081] In conjunction with the second aspect, in certain implementations of the second aspect, the system further includes: a process of supplementing the original fall behavior dataset using Vincent video technology in the data processing module 11: adding multiple scenarios to the original fall behavior dataset, and also taking into account factors such as the age, gender, clothing, and scene location of the elderly at home, and using a scene recognition model to verify the feasibility of the supplemented dataset, and finally obtaining a supplemented fall behavior dataset.

[0082] Alternatively, the process of preprocessing the supplementary fall behavior dataset in the data processing module 11 includes: dividing the video data into positive and negative sample sets according to whether a fall behavior occurs in the video clip, marking corresponding category labels on the positive and negative sample sets respectively, and using a scene recognition model to evaluate the practicality of the video data of the supplementary dataset, and then performing ASCII art encoding processing on the expanded video data based on ASCII art technology, and finally obtaining visually hidden video data with a visual privacy protection effect for the video frame image;

[0083] Alternatively, the calculation process of performing ASCII art encoding processing on the expanded video data based on the ASCII art technology in the data processing module 11 is as follows:

[0084] First, grayscale is performed. The color of each pixel in the color image is determined by three components: Red, Green, and Blue. Each component has 255 possible median values. When these three components are the same, the desired grayscale image X is obtained. Then, the grayscale image X of M×N pixels is divided into n identical m×n small blocks according to the proportion of the selected font; that is, X=[x1,x2,...x n ], each such block will be output as a character, for each pixel block, select a character, the average value of the pixel block of its gray value Y i A good approximation of , and then a block average operation is performed on each pixel block as follows:

[0085]

[0086] where Z i As the sum of the i-th pixel block of image X, is the spatial size of the i-th pixel of image X, Y i is the average value of the i-th pixel block of the grayscale image X;

[0087] Preferably, the privacy protection module 12 extracts image features from the supplementary fall behavior dataset by extracting significant structural features SSF and dense color features DCF;

[0088] Preferably, the process of supplementing the missing features in the hidden state video data that produces the visual privacy protection effect for the video frame image in the fall detection module 13 is:

[0089] The SYM-Track framework is used to accurately and effectively segment and track the main objects in the visual and latent video data with greatly reduced image quality, and to extract the spatiotemporal characteristics of the motion trajectory of the main objects in the visual and latent video frames;

[0090] Preferably, the fall detection module 13 inputs the processed visual and hidden video data into a pre-established YOLO+ posture estimation model and outputs the fall behavior detection result:

[0091] YOLOv7 is used for target detection, and the detected target box information is then input into the Pose model for pose estimation to obtain accurate target detection and pose estimation results. The YOLOv7+ pose estimation model used is based on a deep convolutional neural network architecture. It accepts input images and generates feature maps, which are then used to predict the key point positions of each person. After tracking the target in the ASCII art-encoded fall video, fall detection is performed on the processed visual and latent video data to obtain fall behavior detection results.

[0092] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0093] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium having a computer program stored thereon, which executes the above method when executed by a processor. The storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, device or component, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component.

[0094] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0095] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present disclosure. Various changes and improvements may be made to the present disclosure without departing from the spirit and scope of the present disclosure, and such changes and improvements shall fall within the scope of the present disclosure.

Claims

1. A method for implementing ASCII art visual privacy protection and fall detection, characterized in that: The method comprises the following steps: Receiving an original fall behavior dataset, supplementing the original fall behavior dataset using Vincent video technology to obtain a supplemented fall behavior dataset, and preprocessing the supplemented fall behavior dataset to obtain visually hidden video data with a visual privacy protection effect on video frame images; The process of preprocessing the supplementary fall behavior dataset includes: dividing the video data into positive and negative sample sets according to whether a fall behavior occurs in the video clip, marking corresponding category labels on the positive and negative sample sets, and using a scene recognition model to evaluate the practicality of the video data in the supplementary dataset. Then, ASCII art encoding is performed on the expanded video data based on ASCII art technology, and finally, visual hidden state video data with a visual privacy protection effect is obtained for the video frame images; The calculation process of performing ASCII art encoding processing on the expanded video data based on the ASCII art technology is as follows: First, grayscale is performed. The color of each pixel in the color image is determined by three components: Red, Green, and Blue. Each component has 255 possible median values. When these three components are the same, the desired grayscale image X is obtained. Then, the grayscale image X of M×N pixels is divided into n identical m×n small blocks according to the proportion of the selected font; that is, X=[x1,x2,...x n ], each such block will be output as a character, for each pixel block, select a character, the average value of the pixel block of its gray value Y i A good approximation of , and then a block average operation is performed on each pixel block as follows: where Z i As the sum of the i-th pixel block of image X, is the spatial size of the i-th pixel of image X, Y i is the average value of the i-th pixel block of the grayscale image X; Extract image features from the supplementary fall behavior dataset, input the image features into a pre-established support vector regression model, and output the image privacy protection score; The lost features in the visual and latent video data that produce the visual privacy protection effect of the video frame image are supplemented to obtain the processed visual and latent video data, which is input into the pre-established YOLO+ posture estimation model to output the fall behavior detection result.

2. The method for implementing ASCII art visual privacy protection and fall detection according to claim 1, wherein: The process of using Vincent video technology to supplement the original fall behavior dataset is as follows: multiple scenarios are added to the original fall behavior dataset, and the age, gender, clothing, and scene location factors of the elderly at home are also taken into consideration. The feasibility of the supplemented dataset is verified by using a scene recognition model, and finally a supplemented fall behavior dataset is obtained.

3. The method for implementing ASCII art visual privacy protection and fall detection according to claim 1, wherein: The extraction method supplements the image features in the fall behavior dataset by extracting significant structural features (SSF) and dense color features (DCF).

4. The method for implementing ASCII art visual privacy protection and fall detection according to claim 1, wherein: The process of supplementing the lost features in the hidden state video data that produces the visual privacy protection effect on the video frame image: The SYM-Track framework is used to accurately and effectively segment and track the main objects in the visual and latent video data with greatly reduced image quality, and to extract the spatiotemporal characteristics of the motion trajectory of the main objects in the visual and latent video frames.

5. The method for implementing ASCII art visual privacy protection and fall detection according to claim 1, wherein: The process of inputting the processed visual and hidden video data into the pre-established YOLO+ posture estimation model and outputting the fall behavior detection result: YOLOv7 is used for target detection, and the detected target box information is then input into the Pose model for pose estimation to obtain accurate target detection and pose estimation results. The YOLOv7+ pose estimation model used is based on a deep convolutional neural network architecture. It accepts input images and generates feature maps, which are then used to predict the key point positions of each person. After tracking the target in the ASCII art-encoded fall video, fall detection is performed on the processed visual and latent video data to obtain fall behavior detection results.

6. An ASCII art visual privacy protection and fall detection implementation system, characterized in that, include: a data processing module, configured to receive an original fall behavior dataset, supplement the original fall behavior dataset using Vincent video technology to obtain a supplemented fall behavior dataset, and preprocess the supplemented fall behavior dataset to obtain visually hidden video data that produces a visual privacy protection effect for video frame images; The preprocessing process of the supplementary fall behavior dataset in the data processing module includes: dividing the video data into positive and negative sample sets based on whether a fall occurred in the video clip, marking the positive and negative sample sets with corresponding category labels, and using a scene recognition model to evaluate the practicality of the supplementary dataset video data. The expanded video data is then ASCII art encoded based on ASCII art technology, ultimately obtaining visually hidden video data with visual privacy protection effect on the video frame images; The calculation process of ASCII art encoding of the expanded video data based on ASCII art technology in the data processing module is as follows: First, grayscale is performed. The color of each pixel in the color image is determined by three components: Red, Green, and Blue. Each component has 255 possible median values. When these three components are the same, the desired grayscale image X is obtained. Then, the grayscale image X of M×N pixels is divided into n identical m×n small blocks according to the proportion of the selected font; that is, X=[x1,x2,...x n ], each such block will be output as a character, for each pixel block, select a character, the average value of the pixel block of its gray value Y i A good approximation of , and then a block average operation is performed on each pixel block as follows: where Z i As the sum of the i-th pixel block of image X, is the spatial size of the i-th pixel of image X, Y i is the average value of the i-th pixel block of the grayscale image X; The privacy protection module is used to extract image features from the supplementary fall behavior dataset, input the image features into a pre-established support vector regression model, and output an image privacy protection score; The fall detection module is used to supplement the lost features in the visual and latent video data that produce the visual privacy protection effect of the video frame image, obtain the processed visual and latent video data, input the processed visual and latent video data into the pre-established YOLO+ posture estimation model, and output the fall behavior detection result.

7. The ASCII art visual privacy protection and fall detection implementation system according to claim 6, characterized in that: The data processing module uses Vincent video technology to supplement the original fall behavior dataset: multiple scenarios are added to the original fall behavior dataset, and the age, gender, clothing, and scene location factors of the elderly at home are also taken into account. The feasibility of the supplemented dataset is verified using a scene recognition model, and finally a supplemented fall behavior dataset is obtained; The privacy protection module extracts image features from the supplementary fall behavior dataset by extracting significant structural features (SSF) and dense color features (DCF); The process of supplementing the missing features in the hidden video data to produce visual privacy protection effects on the video frame images in the fall detection module: The SYM-Track framework is used to accurately and effectively segment and track the main objects in the visual and latent video data with greatly reduced image quality, and to extract the spatiotemporal characteristics of the motion trajectory of the main objects in the visual and latent video frames; The fall detection module inputs the processed visual and hidden video data into the pre-established YOLO+ posture estimation model and outputs the fall behavior detection results: YOLOv7 is used for target detection, and the detected target box information is then input into the Pose model for pose estimation to obtain accurate target detection and pose estimation results. The YOLOv7+ pose estimation model used is based on a deep convolutional neural network architecture. It accepts input images and generates feature maps, which are then used to predict the key point positions of each person. After tracking the target in the ASCII art-encoded fall video, fall detection is performed on the processed visual and latent video data to obtain fall behavior detection results.

8. A terminal device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: The memory stores a computer program that can be run on a processor. When the processor loads and executes the computer program, it adopts an ASCII art visual privacy protection and fall detection implementation method according to any one of claims 1 to 5.

Images