In the process of real-time detection of pig body temperature based on thermal infrared video, a method for accurately detecting the ear root temperature of pigs was proposed to solve the problem that the head posture of pigs cannot accurately detect the ear root temperature. Firstly, according to the pig head movement trajectory data, the feeding pen channel was divided into the best ear root temperature measurement area; then, a position offset algorithm was proposed to detect the head posture correction frame in the best ear root temperature measurement area (head posture correct frame, HPCF). Finally, a pig head and ear root detection model based on YOLO v4 was constructed, and the pig head and ear root area were accurately positioned to realize the automatic detection of HPCF, and extract the left and right ear root detection of HPCF respectively. The highest temperature in the frame was taken as the root temperature of each ear. The test results showed that the average detection accuracy (mAP) based on the YOLO v4 model reached 93.15%, and the head and ear roots were positioned accurately; the HPCF detection accuracy rate were 91.33%; the ear root temperature extracted by the algorithm and the manually extracted ear root temperature were compared, and the results were analyzed. It showed that for the temperature of the left and right ear roots in the tested HPCF images, the errors of 97% and 98% of the test images were within 0.3℃. The above research results can provide technical means for real-time monitoring and early warning of abnormal body temperature of pigs.