Edge-distributed fusion of camera-LiDAR for robust moving object localization

Object localization plays a crucial role in computational perception, enabling applications ranging from surveillance to autonomous navigation. This can be leveraged by fusing data cameras and LiDARs (Light Detection Ranging). However, there are challenges employing current fusion methods edge devices, while keeping the process flexible modular. paper presents method for multiple object that fuses LiDAR camera with low-latency, flexibility scalability. Data is obtained 360° surround view four scanning distributed over embedded devices. The proposed technique (a) discriminates dynamic objects scene raw point clouds, clusters their respective points obtain compact representation 3D space (b) asynchronously fuse centroids detection neural networks each detection, localization, tracking. meets above functionalities low-latency increased field of safer navigation, even intermittent flow labels bounding boxes models. That makes our system distributed, modular, scalable agnostic model, distinguishing it state-of-art. Finally, implemented validated both indoor environment publicly available outdoor KITTI 360 set. occurs much faster accurate when compared traditional non-data driven latency competitive other non-embedded deep learning methods. mean error estimated ≈ 5 cm precision 2 navigation 15 m (error percentage 0.3%). Similarly, 30 3 35 on set 0.8%).