Thermal image-based hybrid object detection method using yolo model and object tracking technique

A thermal image-based hybrid object detection method using a YOLO model and an object tracking technique, the thermal image-based hybrid object detection method comprising:

• a) using a thermal image camera, obtaining a plurality of thermal image data, wherein the plurality of thermal image data comprises a first thermal image data and a second thermal image data;
• b) using an image processor, extracting an object region from the first thermal image data of the thermal image camera with an YOLO model installed on the image processor and creating a first object region coordinate data by returning coordinates of the object region;
• c) using an output device, outputting a first image reflecting the first object region coordinate data received from the image processor when the output device requests transmission of the first image;
• d) using a CPU-based AI object detector, creating a second object region coordinate data by tracking multiple object regions from the second thermal image data of the thermal image camera using the first object region coordinate data created by the image processor and by returning coordinates of the multiple object regions;
• and
• e) using the output device, outputting a second image reflecting the second object region coordinate data received from the CPU-based AI object detector wherein the step of b) comprises:
• b-1) adjusting a size of the first thermal image data received from the thermal image camera using a first preprocessor;
• b-2) augmenting the first thermal image data preprocessed by the first preprocessor into a plurality of thermal image data through shade adjustment and regularization using an augmenter;
• b-3) adjusting a size of the plurality of thermal image data augmented through the YOLO model into a size equivalent to a size of the first thermal image data obtained by the thermal image camera and extracting one or more object regions from the plurality of augmented thermal image data using an object region extractor;
• and
• b-4) creating the first object region coordinate data by adjusting and returning coordinates of the one or more object regions configured in the plurality of thermal image data,
• wherein the step of b-2) comprises augmenting the preprocessed first thermal image data into at least two thermal image data of different brightness.