Accurate 3D perception is essential for autonomous driving. Addressing the environmental perception requirements of autonomous driving, we have conducted research into omnidirectional depth estimation algorithm. Building upon this foundation, we have further explored integrated sensing and computing algorithm deployed on hardware, as well as occupancy network aided by depth information.

Depth Estimation

We have studied depth estimation from multi-camera systems to obtain structural information of the surrounding environment for autonomous driving systems. Here is a demo video. More details are available in this page

Integrated Sensing and Computing

We propose an omnidirectional depth estimation system based on near-sensor computing architecture. The proposed work achieves load balancing by task partitioning, while reducing transmission bandwidth through the feature projection and learnable codec. More details are available in this page

Occupancy Network

Based on the depth information provided by our lab's Depth Estimation Network, we propose a Sketch-Coloring framework based on cylindrical voxel. Here is a demo video. Experimental results demonstrate that our Sketch-Coloring network significantly enhances 3D perception performance, especially in nearby regions, which makes our method a promising solution for autonomous driving perception. More details are available in this page