Alioscia Petrelli

Traffic sign detection via interest region extraction

Mobile mapping systems acquire massive amount of data under uncontrolled conditions and pose new challenges to the development of robust computer vision algorithms. In this work, we show how a combination of solid image analysis and pattern recognition techniques can be used to tackle the problem of traffic sign detection in mobile mapping data. Different from the majority of existing systems, our pipeline is based on interest regions extraction rather than sliding window detection. Thanks to the robustness of local features, the proposed pipeline can withstand great appearance variations, which typically occur in outdoor data, especially dramatic illumination and scale changes. The proposed approach has been specialized and tested in three variants, each aimed at detecting one of the three categories of mandatory, prohibitory and danger traffic signs, according to the experimental setup of the recent German Traffic Sign Detection Benchmark competition. Besides achieving very good performance in the on-line competition, our proposal has been successfully evaluated on a novel, more challenging dataset of Italian signs, thereby proving its robustness and suitability to automatic analysis of real-world mobile mapping data. HighlightsThe paper presents in detail the design of a Traffic Sign Detection pipeline.Interest regions are an effective tool to feed a Traffic Sign Detection pipeline.A context-aware and a traffic light filter can effectively prune false positives.Our algorithm obtains competitive results on a public benchmark dataset.Our pipeline achieves promising results on a challenging mobile mapping dataset.

A Repeatable and Efficient Canonical Reference for Surface Matching

The paper investigates on canonical references used for local surface description and matching. We formulate a novel proposal and carry out an extensive experimental evaluation addressing two major surface matching scenarios, namely shape registration and object recognition. We provide also a methodological contribution as, unlike previous work in the field, we propose a repeatability metric that captures the actual impact of the adopted local reference frame algorithm within surface matching tasks based on local 3D descriptors. Our proposal outperforms existing algorithms by a wide margin on several datasets acquired with different devices, such as laser scanners, stereo cameras and the Kinect, and in experiments relying on randomly extracted feature as well as state-of-the art key points.

A traffic sign detection pipeline based on interest region extraction

In this paper we present a pipeline for automatic detection of traffic signs in images. The proposed system can deal with high appearance variations, which typically occur in traffic sign recognition applications, especially with strong illumination changes and dramatic scale changes. Unlike most existing systems, our pipeline is based on interest regions extraction rather than a sliding window detection scheme. The proposed approach has been specialized and tested in three variants, each aimed at detecting one of the three categories of Mandatory, Prohibitory and Danger traffic signs. Our proposal has been evaluated experimentally within the German Traffic Sign Detection Benchmark competition.

On the Affinity between 3D Detectors and Descriptors

The literature on local invariant 3D features is growing, also fostered by the advent of cheap off-the-shelf 3D sensors. Although several recent proposals in the field include both a detector and a descriptor, some of the most successful and used descriptors do not define a companion detector. Moreover, as vouched by the related field of image features, detectors and descriptors defined within the same proposal do not necessarily yield the highest performance when used together. Hence, in this work we investigate on the effectiveness of the many possible combinations between state-of-the-art 3D detectors and descriptors, so as to identify optimal pairs as well as highlight well-matched detectors for those descriptors lacking a companion feature detection algorithm.

Automatic detection of pole-like structures in 3D urban environments.

This work aims at automatic detection of man-made pole-like structures in scans of urban environments acquired by a 3D sensor mounted on top a moving vehicle. Pole-like structures, such as e.g. road signs and streetlights, are widespread in these environments, and their reliable detection is relevant to applications dealing with autonomous navigation, facility damage detection, city planning and maintenance. Yet, due to the characteristic thin shape, detection of man-made pole-like structures is significantly prone to both noise as well as occlusions and clutter, the latter being pervasive nuisances when scanning urban environments. Our approach is based on a “local” stage, whereby local features are classified and clustered together, followed by a “global” stage aimed at further classification of candidate entities. The proposed pipeline turns out effective in experiments on a standard publicly available dataset as well as on a challenging dataset acquired during the project for validation purposes.

Pairwise Registration by Local Orientation Cues

Inspired by recent work on robust and fast computation of 3D Local Reference Frames (LRFs), we propose a novel pipeline for coarse registration of 3D point clouds. Key to the method are: (i) the observation that any two corresponding points endowed with an LRF provide a hypothesis on the rigid motion between two views, (ii) the intuition that feature points can be matched based solely on cues directly derived from the computation of the LRF, (iii) a feature detection approach relying on a saliency criterion which captures the ability to establish an LRF repeatably. Unlike related work in literature, we also propose a comprehensive experimental evaluation based on diverse kinds of data (such as those acquired by laser scanners, Kinect and stereo cameras) as well as on quantitative comparison with respect to other methods. We also address the issue of setting the many parameters that characterize coarse registration pipelines fairly and realistically. The experimental evaluation vouches that our method can handle effectively data acquired by different sensors and is remarkably fast.

Analysis of Compact Features for RGB-D Visual Search

Anticipating the oncoming integration of depth sensing into mobile devices, we experimentally compare different compact features for representing RGB-D images in mobile visual search. Experiments on 3 state-of-the-art datasets, addressing both category and instance recognition, show how Deep Features provided by Convolutional Neural Networks better represent appearance information, whereas shape is more effectively encoded through Kernel Descriptors. Moreover, our evaluation suggests that learning to weight the relative contribution of depth and appearance is key to deploy effectively depth sensing in forthcoming mobile visual search scenarios.

RGB-D Visual Search with Compact Binary Codes

As integration of depth sensing into mobile devices is likely forthcoming, we investigate on merging appearance and shape information for mobile visual search. Accordingly, we propose an RGB-D search engine architecture that can attain high recognition rates with peculiarly moderate bandwidth requirements. Our experiments include a comparison to the CDVS (Compact Descriptors for Visual Search) pipeline, candidate to become part of the MPEG-7 standard, and contribute to elucidate on the merits and limitations of joint deployment of depth and color in mobile visual search.

Learning to Weight Color and Depth for RGB-D Visual Search.

Both color and depth information may be deployed to seek by content through RGB-D imagery. Previous works dealing with global descriptors for RGB-D images advocate a decision level fusion whereby independently computed color and depth representations are juxtaposed to pursue similarity search. Differently, in this paper we propose a learning-to-rank paradigm aimed at weighting the two information channels according to the specific traits of the task and data at hand, thereby effortlessly addressing the potential diversity across applications. In particular, we propose a novel method, referred to as kNN-rank, which can learn the regularities among the outputs yielded by similarity-based queries. A further novel contribution of this paper concerns the HyperRGBD framework, a set of tools conceived to enable seamless aggregation of existing RGB-D datasets in order to obtain new data featuring desired peculiarities and cardinality.