Fire environment detection method and system based on image data

Abstract

The present invention relates to the field of fire environment detection technology, specifically to a fire environment detection method and system based on image data. In the initialization stage, the system generates a data stream identifier, then builds a fire detection model, and improves the performance and efficiency of the model through the optimized FC2f module and the SharedDetect lightweight detection head. Finally, in the real-time detection stage, the video stream collected by the camera is transmitted to the environmental data processing module, the environmental data processing module preliminarily processes the video stream, the fire environment detection module confirms the source of the video stream, and sends it to the fire detection model for analysis, outputs the detection result, divides the fire risk level, and the decision support module tracks the fire occurrence area and generates emergency management suggestions. The present invention realizes efficient and accurate early detection of the fire environment through a lightweight YOLOv8 model and auditing the source of the video stream, and has the beneficial technical effects of real-time detection, intelligent response and avoiding data disorder.

Description

Technical Field

[0001] The present invention relates to the technical field of fire environment detection, and in particular to a fire environment detection method and system based on image data. Background Art

[0002] With the development of modern society and the acceleration of urbanization, fire accidents occur frequently, posing a huge threat to people's lives and property safety. Although traditional fire detection technologies, such as smoke detectors and temperature sensors, can play a role in fire warning to a certain extent, they often need to reach a certain scale before they can take effect. There are problems such as delayed response, limited coverage, and high false alarm rate.

[0003] In recent years, with the rapid development of computer vision technology and deep learning algorithms, fire detection methods based on image data have gradually become a research hotspot. Through intelligent analysis of fire characteristics such as flames and smoke, image processing technology can detect fire signs earlier and issue warnings in time, thereby reducing the losses caused by fire to people. Compared with traditional fire detection methods, fire detection methods based on image data have the advantages of fast detection speed, strong real-time performance, wide application scenarios, and low false alarm rate. Especially in large areas (such as warehouses, shopping malls, industrial plants, etc.), it has greater application potential;

[0004] The Chinese invention patent with the announcement number CN116453029B discloses a building fire environment detection method based on image data. The method obtains a video image of the building environment, grayscales the image, obtains the importance of the pixels in the pre-processed image, and then obtains an adaptive filter window based on the importance of the pixels to achieve adaptive adjustment of the filter window, effectively avoiding errors caused by manual adjustment based on experience, and then enhances the image quality through the adaptive filter window. Based on the enhanced image, the fire detection result of the building environment is obtained;

[0005] However, in actual production life, blind spots and false alarms frequently occur in fire monitoring, the detection accuracy is low, and the deep learning model needs to process a large number of image frames. The system concurrency requirements are high, and the system data is prone to disorder. It is difficult to support the synchronous processing of multiple video streams, and it is difficult to ensure the real-time and accuracy of detection. Summary of the invention

[0006] The purpose of the present invention is to address the problems existing in the background technology and to propose a fire environment detection method and system based on image data.

[0007] The technical solution of the present invention is a fire environment detection method based on image data, comprising the following specific implementation steps:

[0008] S1. In the system initialization phase, the system administrator deploys cameras in the monitoring area, establishes the association between <camera, monitoring area>, establishes communication connections between system components, and performs initialization parameter settings;

[0009] S2. During the real-time fire environment detection stage of the system, any camera collects the video stream of the monitoring area, and generates a data stream identifier through the digital signal processing chip inside the camera, with timestamp information, and transmits the generated data packet to the environmental data processing module;

[0010] S3, the environmental data processing module receives the data packet, and performs preliminary processing on the acquired video stream, extracts continuous image frames at a fixed frame rate of 10 frames per second, scales each image frame to a standard size, processes each image frame, and then transmits the processed data to the fire environment detection module;

[0011] S4, the fire environment detection module verifies the data source, matches the video stream with the camera that collects the video stream, and uses the constructed lightweight fire detection model for analysis, automatically detects the flame and smoke areas in the image, and transmits the output detection results {bounding box, category label, confidence} to the decision support module;

[0012] S5. The decision support module sets a threshold according to the confidence level, filters out the detection results with low confidence, and divides the fire risk level based on the fire level classification rules, obtains the camera number that collects the fire environment video, and confirms the fire occurrence area according to <camera, monitoring area>. Then, the decision support module generates emergency management suggestions based on the above fire risk level, and automatically implements the corresponding emergency management operations, and then sends the fire registration and emergency management suggestions to the visualization module;

[0013] S6. The visualization module stores {camera number, video, bounding box, category label, confidence, emergency management suggestion, collection time} in the system database and displays it on the display device.

[0014] Preferably, the lightweight fire detection model is built based on the YOLOv8 model, the FasterNet Block module is introduced into the C2f module, combined with partial convolution to form a new FC2f module, the C2f module in YOLOv8 is replaced with the FC2f module, and the original detection head is replaced with the lightweight detection head SharedDetect built based on group normalization and shared convolution technology.

[0015] Preferably, each frame of image is scaled to a standard size of 640×640 pixels.

[0016] Preferably, the system initialization parameters in S1 are set as follows:

[0017] S4-1. Decision support module implementation selects a bilinear mapping e: G1×G1→G T , select a secure hash function: H1: {0, 1} * →Z q * 、H2:{0,1} * ×G1→Z q * ;

[0018] Among them, G1 is the additive group of prime number order q, the generator is g, G T is the multiplicative group of prime order q, {0, 1} * Represents a binary string of arbitrary length, Z q * represents the cyclic group of integer multiplication, i.e., the set of non-zero integers modulo q {1, 2, ..., q-1};

[0019] S4-2. The decision support module selects a random number α∈Z q * , calculate the data processing parameter Pdp = α × g ∈ G1, and publicly verify the parameter set {G1, G T , α, e, g, Pdp, H1, H2};

[0020] S4-3, any camera C i The internal digital signal processing chip selects a random number β∈Z q * , calculate the video stream encoding parameter Pen = β × g, calculate the temporary encoding parameter: TPen1 = H1 (ID i , Pen), , calculate the following auxiliary identification generation parameters: PIg1=Pdp+TPen1×g, PIg2=β×PIg1, and make the parameters {β, PIg2} public.

[0021] Preferably, the implementation process of the digital signal processing chip inside the camera generating the data stream identifier is as follows:

[0022] S5-1. Calculate real-time data stream summary information Ab i =H2(video i , ID i , PIg2);

[0023] Among them, video i It is the binary string form of data stream;

[0024] S5-2. Calculate data flow identifier I i : .

[0025] Preferably, the fire environment detection module verifies the source of the data packet and implements the steps of matching the video stream with the camera that collects the video stream as follows:

[0026] S6-1. Calculate real-time data stream summary information Ab ' i =H2(video i , ID i , PIg2);

[0027] S6-2. Calculation of verification factor FI i =H1(ID i , Pen);

[0028] S6-3, if the equation If established, the video stream is confirmed i ID i Camera C i collection.

[0029] The technical solution of the present invention is a fire environment detection system based on image data, which is used to execute the fire environment detection method based on image data proposed above, comprising:

[0030] The video monitoring module is equipped with several cameras to collect video streams of the monitoring area in real time;

[0031] Environmental data processing module, used to pre-process the video stream to ensure clear and high-quality images in complex environments;

[0032] The fire environment detection module uses a pre-trained fire detection model to extract fire features in the video and identify possible signs of fire in the video frame;

[0033] The decision support module is used to trigger the alarm device when the fire environment detection module identifies fire-related features, send the fire alarm to the monitoring center and relevant personnel, and link the fire-fighting equipment;

[0034] The visualization module displays the collected environmental video stream through the display device and stores the fire detection data in the video stream.

[0035] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0036] The present invention proposes a fire environment detection method and system based on image data, which covers three stages: system initialization, model construction and training, and real-time environment detection. In the initialization stage, cameras are arranged according to the monitoring area and associated with the monitoring area. The DSP chip inside the camera generates a data stream identifier and sets parameters. Then the system collects data sets of fire scenes and non-fire scenes, marks flame and smoke areas, and builds a fire detection model based on the YOLOv8 model. The performance and efficiency of the model are improved by the optimized FC2f module and SharedDetect lightweight detection head. Finally, in the real-time detection stage, the video stream collected by the camera is transmitted to the environmental data processing module, and the environmental data processing module performs preliminary processing on the received video stream. After that, the system audits the received video stream, confirms the source of the video stream, and sends it to the fire detection model for analysis, identifies the fire area, outputs the bounding box, category label and confidence, and judges the fire risk level according to the confidence. The decision support module tracks the fire occurrence area, responds step by step and generates emergency management suggestions. The system displays the results on the visualization module and stores them in the database.

[0037] The present invention achieves efficient and accurate early fire detection by optimizing the YOLOv8 model, lightweight detection module and auditing video stream source, and has the beneficial technical effects of real-time detection, rapid response and avoidance of data disorder. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a system architecture diagram of a fire environment detection system based on image data proposed by the present invention;

[0039] Figure 2 This is the network architecture diagram of the fire detection model;

[0040] Figure 3 This is the network structure diagram of the SPPF module;

[0041] Figure 4 This is the network structure diagram of the CBS module;

[0042] Figure 5 This is the network structure diagram of the lightweight module FC2f;

[0043] Figure 6 This is the network structure diagram of the FasterNet Block module;

[0044] Figure 7 This is a schematic diagram of the structure of the lightweight detection head SharedDetect. DETAILED DESCRIPTION

[0045] Embodiment 1, as Figure 1As shown, a fire environment detection system based on image data proposed by the present invention includes: a video monitoring module, an environment data processing module, a fire environment detection module, a decision support module and a visualization module;

[0046] The video monitoring module is equipped with several cameras to collect video streams of the monitoring area in real time;

[0047] The environmental data processing module is used to pre-process the video stream, including but not limited to denoising, image enhancement, and background modeling, to ensure that clear and high-quality images can be obtained in complex environments;

[0048] The fire environment detection module uses a pre-trained fire detection model to extract fire features in the video and identify possible signs of fire in the video frame;

[0049] The decision support module is used to trigger the alarm device when the fire environment detection module identifies fire-related features, send the fire alarm to the monitoring center or relevant personnel, and link the fire-fighting equipment;

[0050] The visualization module displays the collected environmental video stream through the display device and stores the fire detection data in the video stream to facilitate retrospective analysis after a fire incident occurs.

[0051] Embodiment 2, a fire environment detection method based on image data proposed by the present invention, which is applicable to a fire environment detection system based on image data proposed in Embodiment 1, comprises the following implementation steps:

[0052] System initialization phase:

[0053] S1. Arrange cameras according to the size and shape of the monitoring area to ensure full coverage of the entire area. Cameras should be deployed in key areas (such as near electrical equipment, storage areas, and near flammable items) to ensure key monitoring of high fire risk areas. Use NTP (Network Time Protocol) technology to synchronize camera time.

[0054] S2, establish the association between the camera and the monitoring area: {C1, A1}, {C2, A2}, ..., {C i , A i}, ..., {C n , A n} and store it in the system database;

[0055] Among them, C i Indicates the number is ID i Camera, A i Indicates the monitoring area numbered i;

[0056] S3. The system establishes communication connections between the components of the system and implements the following operations:

[0057] S3-1. The decision support module implements the following operations:

[0058] (1) Select a bilinear mapping e: G1×G1→G T ;

[0059] Among them, G1 is the additive group of prime number order q, the generator is g, G T is the multiplicative group of prime order q;

[0060] (2) Select the following secure hash function:

[0061] H1: {0, 1} * →Z q * ;

[0062] H2: {0, 1} * ×G1→Z q * ;

[0063] Among them, {0, 1} * Represents a binary string of arbitrary length, Z q * represents the cyclic group of integer multiplication, i.e., the set of non-zero integers modulo q {1, 2, ..., q-1};

[0064] (3) Select a random number α∈Z q * , calculate the data processing parameter Pdp=α×g∈G1;

[0065] (4) Public verification parameter group {G1, G T ,e,g,Pdp,H1,H2};

[0066] S3-2, any camera C i The internal digital signal processing chip (DSP) performs the following operations:

[0067] (1) Select a random number β∈Z q * , calculate the video stream encoding parameter Pen=β×g;

[0068] (2) Calculate the following temporary coding parameters:

[0069] TPen1=H1(ID i , Pen);

[0070] ;

[0071] (3) Calculate the following auxiliary identification generation parameters:

[0072] PIg1=Pdp+TPen1×g;

[0073] PIg2 = β × PIg1;

[0074] (4) Make the parameters {β, PIg2} public;

[0075] Fire detection model construction and training phase:

[0076] S1. Collect image and video datasets containing fire scenes and non-fire scenes, covering various flames, smoke, lighting conditions and background interference. The dataset should include different fire situations indoors, outdoors, during the day and at night.

[0077] S2. Label the fire data, mark the areas of flames and smoke, and use data augmentation techniques (including but not limited to rotation, flipping, and brightness adjustment) to expand the images to improve the generalization ability of the model;

[0078] S3, such as Figure 2 As shown in the figure, a fire detection model is built based on the YOLOv8 model. The model mainly consists of three parts: backbone network (Backbone), neck network (Neck) and head network (Head). Its specific structure is as follows:

[0079] (1) Conv module is the convolution module;

[0080] (2) If Figure 3 As shown in the figure, the SPPF (Spatial Pyramid Pooling Fast) module is a key component in YOLOv8 for processing multi-scale feature fusion. It performs pooling operations on feature maps of different scales and splices the results together to enhance the model's detection ability for targets of different sizes.

[0081] (3) The Concat module is a concatenation module, which is used to concatenate feature maps from different layers or branches in the channel dimension;

[0082] (4) Upsample module is an upsampling module used to enlarge the size of the feature map to a larger size;

[0083] (5) If Figure 4As shown in the figure, the CBS module is a basic component, including a convolutional block consisting of a convolutional layer (ConvolutionalLayer, Conv), a batch normalization layer (Batch Normalization Layer, BN) and an activation function layer (ActivationFunction Layer). The activation function used is SiLU:

[0084] ;

[0085] in, For the sigmoid function:

[0086] ;

[0087] (6) Figure 5 As shown in Figure 1, in order to achieve the lightweight of the YOLOv8 model, an improved C2f module (FC2f) is constructed: First, the partial convolution (PConv) mechanism is used to optimize the original C2f module of YOLOv8 to reduce redundant calculations and improve calculation efficiency. Then, the FasterNet Block module (such as Figure 6 As shown in the figure, the C2f module is introduced into the C2f module and combined with PConv to form a new FC2f module. This design effectively integrates the advantages of FasterNet Block, further enhances the feature extraction capability of the model, and reduces the number of parameters and computational complexity. Finally, in the YOLOv8 network, all the original C2f modules are replaced by the optimized FC2f modules, which comprehensively improves the model performance and realizes the lightweight of the network.

[0088] In the figure, the Split operation helps to split the input feature map into multiple sub-feature maps, each of which can then be further processed and fused through different paths. ReLU represents the ReLU activation function;

[0089] (7) Figure 7 As shown in the figure, a lightweight detection head SharedDetect based on group normalization (GroupNorm) and shared convolution technology is constructed. This detection head fully combines the advantages of group normalization and shared convolution. Under the premise of effective fusion of feature information, it greatly reduces the computational burden and complexity. Group normalization can dynamically adjust the normalization strategy according to different feature dimensions, effectively improving the generalization ability of the model, while shared convolution reduces parameter redundancy by sharing weights, making the network structure more compact and the computational efficiency higher. SharedDetect not only improves the detection accuracy, but also achieves efficient reasoning performance in the lightweight model, which is suitable for resource-constrained fire environment detection environments.

[0090] In the figure, the GN_Conv1×1 module represents the group normalization GroupNorm+convolution operation, and the convolution kernel size is 1×1. The GN_Conv3×3 module represents the group normalization GroupNorm+convolution operation, and the convolution kernel size is 3×3. Two GN_Conv3×3 modules share weights. Conv_Box represents the target bounding box prediction convolution module, and three Conv_Box modules share weights. Conv_Cls represents classification convolution, and three Conv_Cls modules share weights. Scale is the scaling factor, which is used to match the detection of targets of different scales.

[0091] S4. Evaluate the model through the validation set, observe the performance of the model in unseen fire scenes, adjust hyperparameters such as learning rate and regularization parameters, improve the detection accuracy of the model, evaluate the model performance through accuracy, recall rate, and F1 value indicators, and ensure that the false alarm rate and missed alarm rate of fire detection are within an acceptable range;

[0092] S5, deploying the trained model to the fire environment detection module;

[0093] Real-time environmental detection stage:

[0094] S1, any camera C i The collected video stream is transmitted to the environmental data processing module through the wireless network. During the data transmission process, the camera C i The internal digital signal processing chip generates a data stream identifier with timestamp information. The implementation process is as follows:

[0095] S1-1. Calculate real-time data stream summary information Ab i =H2(video i , ID i , PIg2);

[0096] Among them, video i It is the binary string form of data stream;

[0097] S1-2. Calculate data flow identifier I i :

[0098] ;

[0099] S1-3, the data packet {I i , video i , T i}Transmit to the environmental data processing module;

[0100] Among them, T i The timestamp format indicating the time when the video stream was captured;

[0101] S2, the environmental data processing module receives the data packet {I i , video i , T i}, and perform preliminary processing on the acquired video stream, the processing process is as follows:

[0102] S2-1, the environmental data processing module extracts continuous image frames at a fixed frame rate of 10 frames per second, scales each frame image to a standard size (640×640 pixels), and normalizes the pixel value to the range of [0, 1] based on normalization processing;

[0103] S2-2 removes noise, adjusts brightness and contrast for each frame to improve the quality of video data;

[0104] S2-3, the processed data packet {ID i , I i , video i , Video Frames(Video i ), T i}Transmit to the fire environment detection module;

[0105] S3, the fire environment detection module inputs each frame of each video stream into the pre-trained fire detection model for analysis, automatically detects the flame and smoke areas in the image, and outputs the possibility of fire. The specific implementation process is as follows:

[0106] S3-1, Fire environment detection module checks data packet {ID i , I i , video i , Video Frames(Video i ), T i}Source, confirm the acquisition source of the video stream, the implementation process is as follows:

[0107] (1) Calculate the real-time data stream summary information Ab ' i =H2(video i , ID i , PIg2);

[0108] (2) Calculation of verification factor FI i =H1(ID i , Pen);

[0109] (3) Calculate the following equation:

[0110] ;

[0111] Eq2=e(g,g);

[0112] (4) If Eq1=Eq2, then confirm video i ID i Camera C i collection;

[0113] S3-2, the fire environment detection module loads the pre-trained fire detection model and inputs each frame image in the video stream into the fire detection model;

[0114] S3-3. The fire detection model analyzes and identifies each frame of the video stream and outputs the following information:

[0115] Bounding Box: represents the coordinates of the fire area detected;

[0116] Class Label: Classify the fire type, including flame or smoke;

[0117] Confidence: represents the confidence of fire detection, between 0 and 1;

[0118] S3-4, the fire environment detection module records the output information of the model, and continuously tracks its position, size and shape, and records the changes of the target in subsequent frames;

[0119] S3-5, the fire environment detection module will detect the result {bounding box Bounding-Box i , Class-Lable i Confidence i}Transmit to decision support module and visualization module;

[0120] S4. As shown in Table 1, the decision support module sets a threshold (0.5) according to the confidence level, filters out the detection results with low confidence, and divides the fire risk level based on the fire level classification rules, and responds step by step;

[0121] Table 1 Fire level classification rules

[0122]

[0123] S5, the decision support module opens the data packet {ID i , I i , video i , T i}, get the collected fire environment video i Camera ID i , according to the association {C i , A i}, confirm the fire area, and then the decision support module generates emergency management advice based on the above fire risk level i , and automatically implement the corresponding emergency management operations, and send the fire level and emergency management suggestions to the visualization module;

[0124] S6, the visualization module will {camera number ID i , video i , Bounding-Box i , Class-Lable i Confidence i , Emergency Management Advise i , T i} is stored in the system database and displayed on the display device.

[0125] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A fire environment detection method based on image data, characterized in that: The specific implementation steps include the following: S1. In the system initialization phase, the system administrator deploys cameras in the monitoring area, establishes the association between <camera, monitoring area>, establishes communication connections between system components, and performs initialization parameter settings; S2. During the real-time fire environment detection stage of the system, any camera collects the video stream of the monitoring area, and generates a data stream identifier through the digital signal processing chip inside the camera, with timestamp information, and transmits the generated data packet to the environmental data processing module; S3, the environmental data processing module receives the data packet, and performs preliminary processing on the acquired video stream, extracts continuous image frames at a fixed frame rate of 10 frames per second, scales each image frame to a standard size, processes each image frame, and then transmits the processed data to the fire environment detection module; S4, the fire environment detection module verifies the data source, matches the video stream with the camera that collects the video stream, and uses the constructed lightweight fire detection model for analysis, automatically detects the flame and smoke areas in the image, and transmits the output detection results {bounding box, category label, confidence} to the decision support module; S5. The decision support module sets a threshold according to the confidence level, filters out the detection results with low confidence, and divides the fire risk level based on the fire level classification rules, obtains the camera number that collects the fire environment video, and confirms the fire occurrence area according to <camera, monitoring area>. Then, the decision support module generates emergency management suggestions based on the above fire risk level, and automatically implements the corresponding emergency management operations, and then sends the fire registration and emergency management suggestions to the visualization module; S6. The visualization module stores {camera number, video, bounding box, category label, confidence, emergency management suggestion, collection time} in the system database and displays it on the display device.

2. A fire environment detection method based on image data according to claim 1, characterized in that: The lightweight fire detection model is built based on the YOLOv8 model. The FasterNet Block module is introduced into the C2f module and combined with partial convolution to form a new FC2f module. The C2f module in YOLOv8 is replaced by the FC2f module, and the original detection head is replaced by the lightweight detection head SharedDetect built based on group normalization and shared convolution technology.

3. The fire environment detection method based on image data according to claim 1, characterized in that: Each frame is scaled to a standard size of 640 × 640 pixels.

4. The fire environment detection method based on image data according to claim 1, characterized in that: The system initialization parameters in S1 are set as follows: S4-1. Decision support module implementation selects a bilinear mapping e: G1×G1→G T , select a secure hash function: H1: {0, 1} * →Z q * 、H2:{0,1} * ×G1→Z q * ; Among them, G1 is the additive group of prime number order q, the generator is g, G T is the multiplicative group of prime order q, {0, 1} * Represents a binary string of arbitrary length, Z q * represents the cyclic group of integer multiplication, i.e., the set of non-zero integers modulo q {1, 2, ..., q-1}; S4-2. The decision support module selects a random number α∈Z q * , calculate the data processing parameter Pdp = α × g ∈ G1, and publicly verify the parameter set {G1, G T , α, e, g, Pdp, H1, H2}; S4-3, any camera C i The internal digital signal processing chip selects a random number β∈Z q * , calculate the video stream encoding parameter Pen = β × g, calculate the temporary encoding parameter: TPen1 = H1 (ID i , Pen), , calculate the following auxiliary identification generation parameters: PIg1=Pdp+TPen1×g, PIg2=β×PIg1, and make the parameters {β, PIg2} public.

5. The method for detecting a fire environment based on image data according to claim 1, characterized in that: The implementation process of the digital signal processing chip inside the camera generating a data stream identifier is as follows: S5-1. Calculate real-time data stream summary information Ab i =H2(video i , ID i , PIg2); Among them, video i It is the binary string form of data stream; S5-2. Calculate data flow identifier I i : .

6. A method for detecting a fire environment based on image data according to claim 1, 4 or 5, characterized in that: The fire environment detection module verifies the source of the data packet and matches the video stream with the camera that collected the video stream. The implementation steps are as follows: S6-1. Calculate real-time data stream summary information Ab ' i =H2(video i , ID i , PIg2); S6-2. Calculation of verification factor FI i =H1(ID i , Pen); S6-3, if the equation If established, the video stream is confirmed i ID i Camera C i collection.

7. A fire environment detection system based on image data, used to execute a fire environment detection method based on image data according to any one of claims 1 to 6, characterized in that: include: The video monitoring module is equipped with several cameras to collect video streams of the monitoring area in real time; Environmental data processing module, used to pre-process the video stream to ensure clear and high-quality images in complex environments; The fire environment detection module uses a pre-trained fire detection model to extract fire features in the video and identify possible signs of fire in the video frame; The decision support module is used to trigger the alarm device when the fire environment detection module identifies fire-related features, send the fire alarm to the monitoring center and relevant personnel, and link the fire-fighting equipment; The visualization module displays the collected environmental video stream through the display device and stores the fire detection data in the video stream.

Images