In a scenario where you need to detect multiple specific objects within images, such as vehicles in a traffic monitoring system, which type of model would you choose and why? Additionally, explain how you would label the images for this specific use case and what metrics would be most relevant for evaluating the model's performance.