Energy efficiency is a crucially raised issue in the transportation sector. Moreover, recent technological advancements have brought attention to autonomous driving systems, which can potentially reduce energy wastage associated with human factors. This study introduces an approach to evaluate deep learning models by considering energy efficiency alongside accuracy, incorporating computation complexity as a key metric. Proposing a flexible method for choosing object detection algorithms based on driving environments, the study highlights the trade-off between accuracy and energy efficiency. The dynamic selection of algorithms is shown to optimize energy consumption in autonomous vehicles, especially in varying driving conditions. Furthermore, the study recommends an Explainable Artificial Intelligence (XAI)-based approach to find a compromise in model accuracy, enhancing transparency and societal acceptance of autonomous systems. Future research can focus on extending the data collection process to a wider range of different environmental conditions, which allows for a more robust comparison of the models and a deeper understanding of the trade-offs.