Urban Intersection Classification: A Comparative Analysis.

Understanding the scene in front of a vehicle is crucial for self-driving vehicles and Advanced Driver Assistance Systems, and in urban scenarios, intersection areas are one of the most critical, concentrating between 20% to 25% of road fatalities. This research presents a thorough investigation on the detection and classification of urban intersections as seen from onboard front-facing cameras. Different methodologies aimed at classifying intersection geometries have been assessed to provide a comprehensive evaluation of state-of-the-art techniques based on Deep Neural Network (DNN) approaches, including single-frame approaches and temporal integration schemes. A detailed analysis of most popular datasets previously used for the application together with a comparison with ad hoc recorded sequences revealed that the performances strongly depend on the field of view of the camera rather than other characteristics or temporal-integrating techniques. Due to the scarcity of training data, a new dataset is created by performing data augmentation from real-world data through a Generative Adversarial Network (GAN) to increase generalizability as well as to test the influence of data quality. Despite being in the relatively early stages, mainly due to the lack of intersection datasets oriented to the problem, an extensive experimental activity has been performed to analyze the individual performance of each proposed systems.

CAPformer: Pedestrian Crossing Action Prediction Using Transformer.

Anticipating pedestrian crossing behavior in urban scenarios is a challenging task for autonomous vehicles. Early this year, a benchmark comprising JAAD and PIE datasets have been released. In the benchmark, several state-of-the-art methods have been ranked. However, most of the ranked temporal models rely on recurrent architectures. In our case, we propose, as far as we are concerned, the first self-attention alternative, based on transformer architecture, which has had enormous success in natural language processing (NLP) and recently in computer vision. Our architecture is composed of various branches which fuse video and kinematic data. The video branch is based on two possible architectures: RubiksNet and TimeSformer. The kinematic branch is based on different configurations of transformer encoder. Several experiments have been performed mainly focusing on pre-processing input data, highlighting problems with two kinematic data sources: pose keypoints and ego-vehicle speed. Our proposed model results are comparable to PCPA, the best performing model in the benchmark reaching an F1 Score of nearly 0.78 against 0.77. Furthermore, by using only bounding box coordinates and image data, our model surpasses PCPA by a larger margin (F1=0.75 vs. F1=0.72). Our model has proven to be a valid alternative to recurrent architectures, providing advantages such as parallelization and whole sequence processing, learning relationships between samples not possible with recurrent architectures.

Vehicle Localization Using 3D Building Models and Point Cloud Matching.

Detecting buildings in the surroundings of an urban vehicle and matching them to building models available on map services is an emerging trend in robotics localization for urban vehicles. In this paper, we present a novel technique, which improves a previous work by detecting building façade, their positions, and finding the correspondences with their 3D models, available in OpenStreetMap. The proposed technique uses segmented point clouds produced using stereo images, processed by a convolutional neural network. The point clouds of the façades are then matched against a reference point cloud, produced extruding the buildings' outlines, which are available on OpenStreetMap (OSM). In order to produce a lane-level localization of the vehicle, the resulting information is then fed into our probabilistic framework, called Road Layout Estimation (RLE). We prove the effectiveness of this proposal, testing it on sequences from the well-known KITTI dataset and comparing the results concerning a basic RLE version without the proposed pipeline.