In the context of climate change and the greenhouse effect, the source and cause of emissions is a well known and much debated issue. This includes the role that agriculture, and in particular livestock production, plays as a source of emissions. Measuring emissions in pig and dairy barns is therefore very important, but also very complex and expensive. This especially applies to free-ventilated barns, which are the preferred choice for animal welfare reasons and have a heterogeneous and dynamic distribution of gas concentrations. An alternative to measuring emissions is to model them using other parameters such as the size and temperature of emission sources. These parameters are mainly surfaces soiled with excrements such as urine or feces and can so far only be measured manually by humans. In this PhD project the research gap in the automated detection of emission sources and emission-relevant parameters will be addressed. For this purpose, both RGB and thermal infrared videos are recorded in several pig and dairy barns over a period of several weeks. The video data is then cut into frames, pre-processed and annotated. Subsequently, AI-based object detection models will be trained on the annotated data for different use cases. The focus is on real-time detection models such as YOLO and RT-DETR. Further the combined use of RGB and IR data modalities through fusion models and image fusion in pre-processing will be investigated. In addition to the ability to model emission dynamics more precisely, the automated detection contributes to an improved farm management in the barn and paves the way for a completely self-organized management in the future.