The increase in the annual quantity of appliance batteries placed on the market shows the drive to develop battery-powered appliances. Due to the high resource consumption and recycling potential, batteries are the focus of legal amendments. Despite the provision of collection infrastructures, batteries are increasingly finding their way into various waste streams. Incorrectly disposed batteries not only endanger the economical treatment of waste, but also endanger the environment and the operating personnel in the event of a fire. These challenges require the development of systems for the detection of batteries in waste streams in order to minimise the risk for the sustainable management of waste. However, the development of reliable systems requires the provision of comprehensive, error-free data sets. In this context, the objective of the present scientific study is to develop basic knowledge about influences on optical detection and data processing. To elucidate various influences on the optical detection of batteries, a battery database was created by recording qualitatively tested waste batteries in flight. The influence of camera settings and component arrangement on image quality was investigated in series of experiments. The variation of the training parameters allowed the analysis of the influences on the classification of batteries by means of artificial neural networks. The training of a YOLOv8 model and the prediction of batteries in artificially vaccinated waste streams provided insights into the applicability of an AI-supported battery detection system. The results of this scientific work show the possibility of detecting batteries in waste streams and form the basis for further optimizations of a detection algorithm.