FESTA: FPGA-Enabled Ground Segmentation Technique for Automotive LiDAR

The automotive industry keeps moving fast toward the development of smarter, safer, and more sustainable autonomous vehicles. Today, these come equipped with advanced driver assistance systems (ADAS), which include sophisticated perception technologies to safely navigate the environment. One of the key sensors present in the perception system is light detection and ranging (LiDAR). It can accurately measure distances to objects and create detailed real-time 3-D maps of the surrounding environment, including obstacles and road boundaries. Extracting the road information to identify the drivable area is one of the most important steps applied to a LiDAR output; however, due to the amount of data a high-resolution sensor generates, this task becomes quite challenging. This article proposes FESTA, a ground segmentation technique accelerated in field-programmable gate arrays (FPGAs) using the ALFA framework, that can execute the ground segmentation step applied to the sensor output in real time. The performed evaluation shows that FESTA requires, on average, 8.92 ms for processing a point cloud frame from a VLP-16 sensor, 14.41 ms for an HDL-32, 40.87 ms for an HDL-64, and 70.59 ms for a VLS-128, while outperforming state-of-the-art algorithms in other performance metrics.