Text this: Enhancing semantic segmentation for autonomous vehicle scene understanding in indian context using modified CANet model