Mirror Target YOLO: An Improved YOLOv8 Method with Indirect Vision for Heritage Buildings Fire Detection

Abstract

Fires can cause severe damage to heritage buildings, making timely fire detection essential. Traditional dense cabling and drilling can harm these structures, so reducing the number of cameras to minimize such impact is challenging. Additionally, avoiding false alarms due to noise sensitivity and preserving the expertise of managers in fire-prone areas is crucial. To address these needs, we propose a fire detection method based on indirect vision, called Mirror Target YOLO (MITA-YOLO). MITA-YOLO integrates indirect vision deployment and an enhanced detection module. It uses mirror angles to achieve indirect views, solving issues with limited visibility in irregular spaces and aligning each indirect view with the target monitoring area. The Target-Mask module is designed to automatically identify and isolate the indirect vision areas in each image, filtering out non-target areas. This enables the model to inherit managers' expertise in assessing fire-risk zones, improving focus and resistance to interference in fire detection.In our experiments, we created an 800-image fire dataset with indirect vision. Results show that MITA-YOLO significantly reduces camera requirements while achieving superior detection performance compared to other mainstream models.

Figures (8)

• Figure 1: When the camera is deployed at a position directly facing the gate and only direct vision is used, due to the irregular shape of the indoor space, the direct vision of the camera is blocked. Among the four plants placed in this space, only two can be observed, and two plants are invisible in the occluded area.
• Figure 2: After the mirrors were deployed, the corresponding target areas could be seen in the indirect vision in the mirrors, realizing the expansion of the camera's field of vision coverage and the alignment of the indirect vision and the target detection area.
• Figure 3: First, deploy the mirror system in the irregular space. Then, input the detection images obtained by the camera into the improved detection model. After the original images pass through the Target-Mask module, only the indirect vision area is retained. The processed images are then pushed into the subsequent modules of the detection model. Finally, the results of targeted detection only for the indirect vision area are obtained.
• Figure 4: The Target-Mask is added to the neck between the data input and the backbone network of YOLOv8. When the image data passes through the Target-Mask module, the module will use the built-in network to find the area location information of each indirect vision in the image, and then use the obtained location information to generate a mask with corresponding area boundary information. Then, the mask is mapped back to the original image. After the mapping processing, the passed image only retains the pixels of the target area and filters out other non-interest areas. Then, the optimized image data is sent to the subsequent network of YOLOv8.
• Figure 5: Example of experimental data.