Recently, on nuScenes, an authoritative evaluation set for autonomous driving, the radar and image pre-fusion algorithm PAI3D proposed by the Beijing Economic and Technological Development Area (BDA)'s enterprise autonomous driving team ranked first around the globe in multi-sensor fusion 3D target detection.

Schematic diagram of radar and image pre-fusion algorithm PAI3D [Photo/kfqgw.beijing.gov.cn]

The nuScenes data set is one of the most widely used public data sets in the field of autonomous driving, and it is also the most authoritative pure visual 3D object detection evaluation set for autonomous driving. 

In the current autonomous driving L4 scheme, perception mainly relies on lidar, but lidar has some defects, such as sparse distance information and no color information. 

Image information can make up for these deficiencies, but it also lacks depth information. Therefore, in technology for autonomous driving, how to use multimodal sensor information to design 3D detection algorithms is the core basis of the autonomous driving perception system.

PAI3D is a 3D target detection algorithm proposed by the company based on its experience in terminal delivery scenarios. 

The algorithm fully considers recognition accuracy, calculation efficiency, sensor calibration errors and other issues. In the actual use of the vehicle, it effectively solves the problem of unstable recognition caused by sparse point clouds in the distance, improves the accuracy of obstacle location and category estimation, reduces the false detection and missed detection of obstacles, and supports the large-scale operation of smart delivery vehicles.

"PAI3D's achievements are of great significance for our company's terminal delivery autonomous driving technology, which means that our terminal delivery autonomous driving technology is already at the world's leading level," said a company staff member. 

Driven by long-term technical investment and innovation, the core competitiveness of the enterprise's three-in-one supply chain logistics technology of software, hardware and system integration has been continuously upgraded. 

Currently, it has formed technology products and solutions for various key links in the supply chain covering parks, warehousing, sorting, transportation, and distribution.