Behavior recognition and tracking of group-housed pigs is the key technology to monitor the pigs’ health in smart farming. In real farming scenarios, the pigs’ overlapping occlusion and illumination change make it still challenging to automatically track the behavior of group-housed pigs. An improved ByteTrack algorithm of behavior tracking was proposed based on YOLOX-X for pig behavior recognition and stable tracking to avoid influence due to the complex scene of pig overlap and occlusion. The algorithm improvement included two parts. One was that the trajectory interpolation post-processing strategy based on BYTE data association was designed and implemented to improve the tracking performance. This improvement reduced the error IDs caused by occlusion and enhanced the stability of tracking. The other was to design a detection anchor frame suitable for group-housed pigs and introduce the behavior category information in the YOLOX-X detection algorithm to realize the behavior tracking of group-housed pigs.The experimental results showed that the improved ByteTrack algorithm achieved a favorable performance with MOTA of 96.1%, IDF1 of 94.5%, IDs of 9 and MOTP of 0.189. Compared with the basic ByteTrack, DeepSORT and JDE methods, it had a significant improvement in MOTA and IDF1, and effectively reduced IDs, which showed that the improved ByteTrack algorithm was able to achieve behavior tracking of grouphoused pigs with stable ID tracking. The method can provide technical support for automatic monitoring of pigs with no contact.