# US 11,908,298 B2 · Smoke detection system and smoke detection method > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,908,298 B2 - Document ID: US11908298B2 - Status: Granted - Kind: B2 - Application number: 17/739,225 - Filed: 2022-05-09 - Published: 2023-06-08 - Granted: 2024-02-20 - Earliest claimed: 2021-12-08 - Assignees: VIA Technologies, Inc. - Applicants: VIA Technologies, Inc. - Inventors: Jia-yo Hsu; Wei-Chung Cheng - Web page: https://patentgrep.com/patent/US11908298B2 ## CPC - G06T 7/254: involving subtraction of images - G08B 17/125: {by using a video camera to detect fire or smoke} - G06T 7/11: Region-based segmentation - G06T 7/174: involving the use of two or more images - G06V 10/26: Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion - G06V 10/54: relating to texture - G06V 10/56: relating to colour - G06V 20/52: Surveillance or monitoring of activities, e.g. for recognising suspicious objects (recognising microscopic objects G06V20/69) - G06V 20/60: Type of objects - G08B 29/186: {Fuzzy logic; neural networks} - G06T 2207/30232: Surveillance - G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL - G08B: SIGNALLING SYSTEMS, e.g. PERSONAL CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING ## Abstract The present disclosure discloses a smoke detection system and a smoke detection method. The smoke detection system includes a camera, a storage unit, and a processor. The camera acquires a current image and a previous image. The storage unit stores a plurality of modules. The processor is coupled with the camera and executes the plurality of modules. The processor generates a difference image based on the current image and the previous image. The processor inputs the current image and the difference image to a semantic segmentation model so that the semantic segmentation model outputs a smoke confidence map. The smoke confidence map is generated based on whether a current environment is a dark environment or a bright environment. The processor analyzes the smoke confidence map to determine whether a smoke event occurs in the current image. Therefore, a reliable smoke detection function can be achieved. ## Claims ### Claim 1 A smoke detection system, comprising: a camera configured to acquire a current image and a previous image; a storage unit configured to store a plurality of modules; and a processor coupled with the camera and configured to execute the plurality of modules to: generate a difference image based on the current image and the previous image; input the current image and the difference image to a semantic segmentation model so that the semantic segmentation model outputs a smoke confidence map, wherein the smoke confidence map is generated based on whether a current environment is a dark environment or a bright environment; and analyze the smoke confidence map to determine whether a smoke event occurs in the current image. ### Claim 2 (depends on 1) The smoke detection system according to claim 1, wherein the camera is configured to sequentially acquire a plurality of images, and the processor is configured to analyze the plurality of images to generate a plurality of smoke confidence maps, wherein the processor is configured to determine whether the smoke event occurs based on the plurality of smoke confidence maps and a smoke alarm fuzzy logic. ### Claim 3 (depends on 1) The smoke detection system according to claim 1, wherein the processor is configured to subtract a plurality of pixel values of the previous image from a plurality of corresponding pixel values of the current image to generate the difference image. ### Claim 4 (depends on 1) The smoke detection system according to claim 1, wherein the processor is configured to determine a number of smoke pixels each having a confidence value greater than a pixel confidence threshold in the smoke confidence map based on the pixel confidence threshold, and the processor is configured to determine whether the number of smoke pixels is greater than a smoke pixel threshold to determine whether the smoke event occurs in the current image. ### Claim 5 (depends on 1) The smoke detection system according to claim 1, wherein the processor is configured to subtract a blue pixel average value from a green pixel average value for the current image or the previous image to obtain a pixel difference value, in response to the processor determining that the pixel difference value is greater than or equal to a pixel threshold, the processor determines that the current environment is the dark environment, and In response to the processor determining that the pixel difference value is less than the pixel threshold, the processor determines that the current environment is the bright environment. ### Claim 6 (depends on 1) The smoke detection system according to claim 1, wherein in response to the processor determining that the current environment is the bright environment, the processor performs image dark channel calculations on the current image and the previous image respectively to generate a current dark channel image and a previous dark channel image, and the processor subtracts a plurality of pixel values of the previous dark channel image from a plurality of corresponding pixel values of the current dark channel image to generate the difference image. ### Claim 7 (depends on 6) The smoke detection system according to claim 6, wherein the processor is configured to perform image erosion processes on the current dark channel image and the previous dark channel image respectively, and then subtract the plurality of pixel values of the previous dark channel image after the image erosion process from the plurality of corresponding pixel values of the current dark channel image after the image erosion process, to generate the difference image. ### Claim 8 (depends on 1) The smoke detection system according to claim 1, wherein in response to the processor determining that the current environment is the dark environment, the processor performs sRGB color space difference operations on the current image and the previous image to obtain the difference image. ### Claim 9 (depends on 1) The smoke detection system according to claim 1, wherein the processor is configured to input a color channel of red pixels in the current image, a color channel of green pixels in the current image, a color channel of blue pixels in the current image, and a gray-scale pixel channel of the difference image as four channels of input data to the semantic segmentation model. ### Claim 10 (depends on 1) The smoke detection system according to claim 1, wherein the processor and the storage unit are provided at a cloud server, and the camera is configured to provide the current image and the previous image to the cloud server through a communication module. ### Claim 11 A smoke detection method, comprising: acquiring a current image and a previous image by a camera; generating, by a processor, a difference image based on the current image and the previous image; inputting, by the processor, the current image and the difference image to a semantic segmentation model so that the semantic segmentation model outputs a smoke confidence map, wherein the smoke confidence map is generated based on whether a current environment is a dark environment or a bright environment; and analyzing, by the processor, the smoke confidence map to determine whether a smoke event occurs in the current image. ### Claim 12 (depends on 11) The smoke detection method according to claim 11, further comprising: acquiring, by the camera, a plurality of images sequentially; analyzing, by the processor, the plurality of images to generate a plurality of smoke confidence maps; and determining, by the processor, whether the smoke event occurs based on the plurality of smoke confidence maps and a smoke alarm fuzzy logic. ### Claim 13 (depends on 11) The smoke detection method according to claim 11, wherein said generating the difference image includes: subtracting, by the processor, a plurality of pixel values of the previous image from a plurality of corresponding pixel values of the current image to generate the difference image. ### Claim 14 (depends on 11) The smoke detection method according to claim 11, wherein said determining whether the smoke event occurs in the current image includes: determining, by the processor, a number of smoke pixels each having a confidence value greater than a pixel confidence threshold in the smoke confidence map based on the pixel confidence threshold; and determining, by the processor, whether the number of smoke pixels is greater than a smoke pixel threshold to determine whether the smoke event occurs in the current image. ### Claim 15 (depends on 11) The smoke detection method according to claim 11, wherein said generating the difference image based on the current image and the previous image includes: subtracting, by the processor, a blue pixel average value from a green pixel average value for the current image or the previous image to obtain a pixel difference value; determining, by the processor, that the current environment is the dark environment in response to the processor determining that the pixel difference value is greater than or equal to a pixel threshold; and determining, by the processor, that the current environment is the bright environment in response to the processor determining that the pixel difference value is less than the pixel threshold. ### Claim 16 (depends on 11) The smoke detection method according to claim 11, wherein said generating the difference image based on the current image and the previous image further includes: in response to the processor determining that the current environment is the bright environment, performing by the processor image dark channel calculations on the current image and the previous image respectively, to generate a current dark channel image and a previous dark channel image; and subtracting, by the processor, a plurality of pixel values of the previous dark channel image from a plurality of corresponding pixel values of the current dark channel image to generate the difference image. ### Claim 17 (depends on 16) The smoke detection method according to claim 16, wherein said generating the difference image based on the current image and the previous image further includes: by the processor, performing image erosion processes on the current dark channel image and the previous dark channel image respectively, and then subtracting the plurality of pixel values of the previous dark channel image after the image erosion process from the plurality of corresponding pixel values of the current dark channel image after the image erosion process to generate the difference image. ### Claim 18 (depends on 11) The smoke detection method according to claim 11, wherein said generating the difference image based on the current image and the previous image further includes: in response to the processor determining that the current environment is the dark environment, performing, by the processor, sRGB color space difference operations on the current image and the previous image to obtain the difference image. ### Claim 19 (depends on 11) The smoke detection method according to claim 11, wherein said inputting the current image and the difference image to the semantic segmentation model further includes: inputting, by the processor, a color channel of red pixels in the current image, a color channel of green pixels in the current image, a color channel of blue pixels in the current image, and a gray-scale pixel channel of the difference image as four channels of input data to the semantic segmentation model. ### Claim 20 (depends on 11) The smoke detection method according to claim 11, wherein the processor and a storage unit storing a plurality of modules to be executed by the processor are provided at a cloud server, and the camera is configured to provide the current image and the previous image to the cloud server through a communication module.