Lorenzo Baraldi

Gesture Recognition in Ego-centric Videos Using Dense Trajectories and Hand Segmentation

We present a novel method for monocular hand gesture recognition in ego-vision scenarios that deals with static and dynamic gestures and can achieve high accuracy results using a few positive samples. Specifically, we use and extend the dense trajectories approach that has been successfully introduced for action recognition. Dense features are extracted around regions selected by a new hand segmentation technique that integrates superpixel classification, temporal and spatial coherence. We extensively testour gesture recognition and segmentation algorithms on public datasets and propose a new dataset shot with a wearable camera. In addition, we demonstrate that our solution can work in near real-time on a wearable device.

A deep multi-level network for saliency prediction

This paper presents a novel deep architecture for saliency prediction. Current state of the art models for saliency prediction employ Fully Convolutional networks that perform a non-linear combination of features extracted from the last convolutional layer to predict saliency maps. We propose an architecture which, instead, combines features extracted at different levels of a Convolutional Neural Network (CNN). Our model is composed of three main blocks: a feature extraction CNN, a feature encoding network, that weights low and high level feature maps, and a prior learning network. We compare our solution with state of the art saliency models on two public benchmarks datasets. Results show that our model outperforms under all evaluation metrics on the SALICON dataset, which is currently the largest public dataset for saliency prediction, and achieves competitive results on the MIT300 benchmark. Code is available at https://github.com/marcellacornia/mlnet.

Hand segmentation for gesture recognition in EGO-vision

Portable devices for first-person camera views will play a central role in future interactive systems. One necessary step for feasible human-computer guided activities is gesture recognition, preceded by a reliable hand segmentation from egocentric vision. In this work we provide a novel hand segmentation algorithm based on Random Forest superpixel classification that integrates light, time and space consistency. We also propose a gesture recognition method based Exemplar SVMs since it requires a only small set of positive sampels, hence it is well suitable for the egocentric video applications. Furthermore, this method is enhanced by using segmented images instead of full frames during test phase. Experimental results show that our hand segmentation algorithm outperforms the state-of-the-art approaches and improves the gesture recognition accuracy on both the publicly available EDSH dataset and our dataset designed for cultural heritage applications.

Gesture Recognition Using Wearable Vision Sensors to Enhance Visitors’ Museum Experiences

We introduce a novel approach to cultural heritage experience: by means of ego-vision embedded devices we develop a system, which offers a more natural and entertaining way of accessing museum knowledge. Our method is based on distributed self-gesture and artwork recognition, and does not need fixed cameras nor radio-frequency identifications sensors. We propose the use of dense trajectories sampled around the hand region to perform self-gesture recognition, understanding the way a user naturally interacts with an artwork, and demonstrate that our approach can benefit from distributed training. We test our algorithms on publicly available data sets and we extend our experiments to both virtual and real museum scenarios, where our method shows robustness when challenged with real-world data. Furthermore, we run an extensive performance analysis on our ARM-based wearable device.

A Deep Siamese Network for Scene Detection in Broadcast Videos

We present a model that automatically divides broadcast videos into coherent scenes by learning a distance measure between shots. Experiments are performed to demonstrate the effectiveness of our approach by comparing our algorithm against recent proposals for automatic scene segmentation. We also propose an improved performance measure that aims to reduce the gap between numerical evaluation and expected results, and propose and release a new benchmark dataset.

Hierarchical Boundary-Aware Neural Encoder for Video Captioning

The use of Recurrent Neural Networks for video captioning has recently gained a lot of attention, since they can be used both to encode the input video and to generate the corresponding description. In this paper, we present a recurrent video encoding scheme which can discover and leverage the hierarchical structure of the video. Unlike the classical encoder-decoder approach, in which a video is encoded continuously by a recurrent layer, we propose a novel LSTM cell which can identify discontinuity points between frames or segments and modify the temporal connections of the encoding layer accordingly. We evaluate our approach on three large-scale datasets: the Montreal Video Annotation dataset, the MPII Movie Description dataset and the Microsoft Video Description Corpus. Experiments show that our approach can discover appropriate hierarchical representations of input videos and improve the state of the art results on movie description datasets.

Shot and Scene Detection via Hierarchical Clustering for Re-using Broadcast Video

Video decomposition techniques are fundamental tools for allowing effective video browsing and re-using. In this work, we consider the problem of segmenting broadcast videos into coherent scenes, and propose a scene detection algorithm based on hierarchical clustering, along with a very fast state-of-the-art shot segmentation approach. Experiments are performed to demonstrate the effectiveness of our algorithms, by comparing against recent proposals for automatic shot and scene segmentation.

Recognizing and Presenting the Storytelling Video Structure With Deep Multimodal Networks

In this paper, we propose a novel scene detection algorithm which employs semantic, visual, textual, and audio cues. We also show how the hierarchical decomposition of the storytelling video structure can improve retrieval results presentation with semantically and aesthetically effective thumbnails. Our method is built upon two advancements of the state of the art: first is semantic feature extraction which builds video-specific concept detectors; and second is multimodal feature embedding learning that maps the feature vector of a shot to a space in which the Euclidean distance has task specific semantic properties. The proposed method is able to decompose the video in annotated temporal segments which allow us for a query specific thumbnail extraction. Extensive experiments are performed on different data sets to demonstrate the effectiveness of our algorithm. An in-depth discussion on how to deal with the subjectivity of the task is conducted and a strategy to overcome the problem is suggested.

Optimized Connected Components Labeling with Pixel Prediction

In this paper we propose a new paradigm for connected components labeling, which employs a general approach to minimize the number of memory accesses, by exploiting the information provided by already seen pixels, removing the need to check them again. The scan phase of our proposed algorithm is ruled by a forest of decision trees connected into a single graph. Every tree derives from a reduction of the complete optimal decision tree. Experimental results demonstrated that on low density images our method is slightly faster than the fastest conventional labeling algorithms.

YACCLAB - Yet Another Connected Components Labeling Benchmark

The problem of labeling the connected components (CCL) of a binary image is well-defined and several proposals have been presented in the past. Since an exact solution to the problem exists and should be mandatory provided as output, algorithms mainly differ on their execution speed. In this paper, we propose and describe YACCLAB, Yet Another Connected Components Labeling Benchmark. Together with a rich and varied dataset, YACCLAB contains an open source platform to test new proposals and to compare them with publicly available competitors. Textual and graphical outputs are automatically generated for three kinds of test, which analyze the methods from different perspectives. The fairness of the comparisons is guaranteed by running on the same system and over the same datasets. Examples of usage and the corresponding comparisons among state-of-the-art techniques are reported to confirm the potentiality of the benchmark.