I believe the most straightforward one would be to create and train your own HAAR Cascade files (one for each object you need to detect) and use them in combination with the ObjectDetector (Custom) VL.OpenCV node. You will need one ObjectDetector for each object type you need detected. Each detector will give you the ROI where the object has been detected in the provided image.
Support for multi-object detection is not yet officially available in the C# opencv wrapper we are currently using (opencvsharp) but it is currently being implemented so it should be supported in the near future.