Present day autonomous vehicle relies on several sensor technologies for it's autonomous functionality. The sensors based on their type and mounted-location on the vehicle, can be categorized as: line of sight and non-line of sight sensors and are responsible for the different level of autonomy. These line of sight sensors are used for the execution of actions related to localization, object detection and the complete environment understanding. The surrounding or environment understanding for an autonomous vehicle can be achieved by segmentation. Several traditional and deep learning related techniques providing semantic segmentation for an input from camera is already available, however with the advancement in the computing processor, the progression is on developing the deep learning application replacing traditional methods. This paper presents an approach to combine the input of camera and lidar for semantic segmentation purpose. The proposed model for outdoor scene segmentation is based on the frustum pointnet, and ResNet which utilizes the 3d point cloud and camera input for the 3d bounding box prediction across the moving and non-moving object and thus finally recognizing and understanding the scenario at the point-cloud or pixel level. For real time application the model is deployed on the RTMaps framework with Bluebox (an embedded platform for autonomous vehicle). The proposed architecture is trained with the CITYScpaes and the KITTI dataset.
A machine learning model can often produce biased outputs for a familiar group or similar sets of classes during inference over an unknown dataset. The generalization of neural networks have been studied to resolve biases, which has also shown improvement in accuracy and performance metrics, such as precision and recall, and refining the dataset's validation set. Data distribution and instances included in test and validation-set play a significant role in improving the generalization of neural networks. For producing an unbiased AI model, it should not only be trained to achieve high accuracy and minimize false positives. The goal should be to prevent the dominance of one class/feature over the other class/feature while calculating weights. This paper investigates state-of-art object detection/classification on AI models using metrics such as selectivity score and cosine similarity. We focus on perception tasks for vehicular edge scenarios, which generally include collaborative tasks and model updates based on weights. The analysis is performed using cases that include the difference in data diversity, the viewpoint of the input class and combinations. Our results show the potential of using cosine similarity, selectivity score and invariance for measuring the training bias, which sheds light on developing unbiased AI models for future vehicular edge services.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.