No abstract

No abstract

Abstract-Collision avoidance is one of the most difficult and challenging automatic driving operations in the domain of intelligent vehicles. In emergency situations, human drivers are more likely to brake than to steer, although the optimal maneuver would, more frequently, be steering alone. This statement suggests the use of automatic steering as a promising solution to avoid accidents in the future. The objective of this paper is to provide a collision avoidance system (CAS) for autonomous vehicles, focusing on pedestrian collision avoidance. The detection component involves a stereo-vision-based pedestrian detection system that provides suitable measurements of the time to collision. The collision avoidance maneuver is performed using fuzzy controllers for the actuators that mimic human behavior and reactions, along with a high-precision Global Positioning System (GPS), which provides the information needed for the autonomous navigation. The proposed system is evaluated in two steps. First, drivers' behavior and sensor accuracy are studied in experiments carried out by manual driving. This study will be used to define the parameters of the second step, in which automatic pedestrian collision avoidance is carried out at speeds of up to 30 km/h. The performed field tests provided encouraging results and proved the viability of the proposed approach.

Abstract-Over the past few years, advanced driver-assistance systems (ADASs) have become a key element in the research and development of intelligent transportation systems (ITSs) and particularly of intelligent vehicles. Many of these systems require accurate global localization information, which has been traditionally performed by the Global Positioning System (GPS), despite its well-known failings, particularly in urban environments. Different solutions have been attempted to bridge the gaps of GPS positioning errors, but they usually require additional expensive sensors. Vision-based algorithms have proved to be capable of tracking the position of a vehicle over long distances using only a sequence of images as input and with no prior knowledge of the environment. This paper describes a full solution to the estimation of the global position of a vehicle in a digital road map by means of visual information alone. Our solution is based on a stereo platform used to estimate the motion trajectory of the ego vehicle and a map-matching algorithm, which will correct the cumulative errors of the vision-based motion information and estimate the global position of the vehicle in a digital road map. We demonstrate our system in large-scale urban experiments reaching high accuracy in the estimation of the global position and allowing for longer GPS blackouts due to both the high accuracy of our visual odometry estimation and the correction of the cumulative error of the mapmatching algorithm. Typically, challenging situations in urban environments such as nonstatic objects or illumination exceeding the dynamic range of the cameras are shown and discussed.

No abstract

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.

No abstract

Fine-grained vehicle classification from images, also known as Vehicle Make and Model Recognition (VMMR), has become an important research topic in the last years, with a growing number of scientific contributions in multiple application areas, such as autonomous vehicles, surveillance systems, traffic monitoring and management, among others. Recent techniques based on deep learning have proven to be very effective in addressing this problem. So effective that, based on the state-of-the-art results (above 95% accuracy), it would seem that the problem is practically solved. However, our main hypothesis is that the existing datasets to date have limited variability, which precludes good and unbiased generalisation of the models trained with them. In particular, it is observed that the test datasets are very similar in nature to those used for training and validation which makes these benchmarks prone to dataset bias and to overfitting. When these systems are tested with more challenging data or data from different datasets performance degrades considerably. In this paper, on the one hand, we evaluate state-of-the-art deep learning models to perform fine-grained vehicle classification and explore multiple training techniques, such as curriculum learning or weighted losses, to mitigate the bias between different makes and models and to assess the limits of current approaches. On the other hand, we analyse the existing datasets, present an additional dataset from a challenging scenario, and merge all the data into a cross-dataset that includes common samples and classes from the existing datasets. In this way, we can evaluate geographical, make and model biases, and performance and generalisation capabilities from a more realistic perspective. The obtained results suggest that we are still far from accurate and unbiased vehicle make and model recognition in realistic traffic and driving scenarios.INDEX TERMS Fine-grained classification, vehicle make and model, dataset bias, curriculum learning, weighted loss, cross-datasets.