Cooperative Perception for 3D Object Detection in Driving Scenarios Using Infrastructure Sensors

3D object detection is a common function within the perception system of an\nautonomous vehicle and outputs a list of 3D bounding boxes around objects of\ninterest. Various 3D object detection methods have relied on fusion of\ndifferent sensor modalities to overcome limitations of individual sensors.\nHowever, occlusion, limited field-of-view and low-point density of the sensor\ndata cannot be reliably and cost-effectively addressed by multi-modal sensing\nfrom a single point of view. Alternatively, cooperative perception incorporates\ninformation from spatially diverse sensors distributed around the environment\nas a way to mitigate these limitations. This article proposes two schemes for\ncooperative 3D object detection using single modality sensors. The early fusion\nscheme combines point clouds from multiple spatially diverse sensing points of\nview before detection. In contrast, the late fusion scheme fuses the\nindependently detected bounding boxes from multiple spatially diverse sensors.\nWe evaluate the performance of both schemes, and their hybrid combination,\nusing a synthetic cooperative dataset created in two complex driving scenarios,\na T-junction and a roundabout. The evaluation shows that the early fusion\napproach outperforms late fusion by a significant margin at the cost of higher\ncommunication bandwidth. The results demonstrate that cooperative perception\ncan recall more than 95% of the objects as opposed to 30% for single-point\nsensing in the most challenging scenario. To provide practical insights into\nthe deployment of such system, we report how the number of sensors and their\nconfiguration impact the detection performance of the system.\n