We introduce multimodal multiresolution filtering of signal intensity, gradient magnitude and orientation channels, in order to capture structure at multiple scales and orientations. A scale correction scheme is proposed in order to achieve scale invariant classification features. To improve recognition we also incorporate 2D and 3D context channels, and deep learning based convolutional channels. Classification is achieved using boosted decision trees over channel features. Finally, we evaluate the proposed methods on multiple benchmarks and present a solution for 360 degree environment perception for an autonomous vehicle."