Are Vision-Language Models Truly Understanding Multi-vision Sensor?