Enrique J. Carmona

A survey of video datasets for human action and activity recognition

Highlights? Description of datasets for video-based human activity and action recognition. ? 68 datasets reported: 28 for heterogeneous and 40 for specific human actions. ? Useful data, such as web for dowloading, published works or ground truth, are provided. ? Datasets are compared and classified. Vision-based human action and activity recognition has an increasing importance among the computer vision community with applications to visual surveillance, video retrieval and human-computer interaction. In recent years, more and more datasets dedicated to human action and activity recognition have been created. The use of these datasets allows us to compare different recognition systems with the same input data. The survey introduced in this paper tries to cover the lack of a complete description of the most important public datasets for video-based human activity and action recognition and to guide researchers in the election of the most suitable dataset for benchmarking their algorithms.

Identification of the optic nerve head with genetic algorithms

OBJECTIVE This work proposes creating an automatic system to locate and segment the optic nerve head (ONH) in eye fundus photographic images using genetic algorithms. METHODS AND MATERIAL Domain knowledge is used to create a set of heuristics that guide the various steps involved in the process. Initially, using an eye fundus colour image as input, a set of hypothesis points was obtained that exhibited geometric properties and intensity levels similar to the ONH contour pixels. Next, a genetic algorithm was used to find an ellipse containing the maximum number of hypothesis points in an offset of its perimeter, considering some constraints. The ellipse thus obtained is the approximation to the ONH. The segmentation method is tested in a sample of 110 eye fundus images, belonging to 55 patients with glaucoma (23.1%) and eye hypertension (76.9%) and random selected from an eye fundus image base belonging to the Ophthalmology Service at Miguel Servet Hospital, Saragossa (Spain). RESULTS AND CONCLUSIONS The results obtained are competitive with those in the literature. The methods generalization capability is reinforced when it is applied to a different image base from the one used in our study and a discrepancy curve is obtained very similar to the one obtained in our image base. In addition, the robustness of the method proposed can be seen in the high percentage of images obtained with a discrepancy delta<5 (96% and 99% in our and a different image base, respectively). The results also confirm the hypothesis that the ONH contour can be properly approached with a non-deformable ellipse. Another important aspect of the method is that it directly provides the parameters characterising the shape of the papilla: lengths of its major and minor axes, its centre of location and its orientation with regard to the horizontal position.

A new video segmentation method of moving objects based on blob-level knowledge

Variants of the background subtraction method are broadly used for the detection of moving objects in video sequences in different applications. In this work we propose a new approach to the background subtraction method which operates in the colour space and manages the colour information in the segmentation process to detect and eliminate noise. This new method is combined with blob-level knowledge associated with different types of blobs that may appear in the foreground. The idea is to process each pixel differently according to the category to which it belongs: real moving objects, shadows, ghosts, reflections, fluctuation or background noise. Thus, the foreground resulting from processing each image frame is refined selectively, applying at each instant the appropriate operator according to the type of noise blob we wish to eliminate. The approach proposed is adaptive, because it allows both the background model and threshold model to be updated. On the one hand, the results obtained confirm the robustness of the method proposed in a wide range of different sequences and, on the other hand, these results underline the importance of handling three colour components in the segmentation process rather than just the one grey-level component.

Solving differential equations with Fourier series and Evolution Strategies

A novel mesh-free approach for solving differential equations based on Evolution Strategies (ESs) is presented. Any structure is assumed in the equations making the process general and suitable for linear and nonlinear ordinary and partial differential equations (ODEs and PDEs), as well as systems of ordinary differential equations (SODEs). Candidate solutions are expressed as partial sums of Fourier series. Taking advantage of the decreasing absolute value of the harmonic coefficients with the harmonic order, several ES steps are performed. Harmonic coefficients are taken into account one by one starting with the lower order ones. Experimental results are reported on several problems extracted from the literature to illustrate the potential of the proposed approach. Two cases (an initial value problem and a boundary condition problem) have been solved using numerical methods and a quantitative comparative is performed. In terms of accuracy and storing requirements the proposed approach outperforms the numerical algorithm.

Using genetic algorithms to improve the thermodynamic efficiency of gas turbines designed by traditional methods

A method for optimizing the thermodynamic efficiency of aeronautical gas turbines designed by classical methods is presented. This method is based in the transformation of the original constrained optimization problem into a new constrained free optimization problem which is solved by a genetic algorithm. Basically, a set of geometric, aerodynamic and acoustic noise constraints must be fulfilled during the optimization process. As a case study, the thermodynamic efficiency of an already optimized by traditional methods real aeronautical low pressure turbine design of 13 rows has been successfully improved, increasing the turbine efficiency by 0.047% and reducing the total number of airfoils by 1.61%. In addition, experimental evidence of a strong correlation between the total number of airfoils and the turbine efficiency has been observed. This result would allow us to use the total number of airfoils as a cheap substitute of the turbine efficiency for a coarse optimization at the first design steps.

On the effect of feedback in multilevel representation spaces for visual surveillance tasks

In this work we propose a general top-down feedback scheme between adjacent description levels to interpret video sequences. This scheme distinguishes two types of feedback: repair-oriented feedback and focus-oriented feedback. With the first it is possible to improve the systems performance and produce more reliable and consistent information, and with the second it is possible to adjust the computational load to match the aims. Finally, the general feedback scheme is used in different examples for a visual surveillance application which improved the final result of each description level by using the information in the higher adjacent level.

Parametric improvement of lateral interaction in accumulative computation in motion-based segmentation

Segmentation of moving objects is an essential component of any vision system. However, its accomplishment is hard due to some challenges such as the occlusion treatment or the detection of objects with deformable appearance. In this paper an artificial neuronal network approach for moving object segmentation, called lateral interaction in accumulative computation (LIAC), which uses accumulative computation and recurrent lateral interaction is revisited. Although the results reported for this approach so far may be considered relevant, the problems faced each time (environment, objects of interest, etc.) make that the system outcome varies. Hence, our aim is to improve segmentation provided by LIAC in a double sense: by removing the detected objects not matching some size or compactness constraints, and by learning suitable parameters that improve the segmentation behavior through a genetic algorithm.

Fast detection of the main anatomical structures in digital retinal images based on intra- and inter-structure relational knowledge

BACKGROUND AND OBJECTIVE The anatomical structure detection in retinal images is an open problem. However, most of the works in the related literature are oriented to the detection of each structure individually or assume the previous detection of a structure which is used as a reference. The objective of this paper is to obtain simultaneous detection of the main retinal structures (optic disc, macula, network of vessels and vascular bundle) in a fast and robust way. METHODS We propose a new methodology oriented to accomplish the mentioned objective. It consists of two stages. In an initial stage, a set of operators is applied to the retinal image. Each operator uses intra-structure relational knowledge in order to produce a set of candidate blobs that belongs to the desired structure. In a second stage, a set of tuples is created, each of which contains a different combination of the candidate blobs. Next, filtering operators, using inter-structure relational knowledge, are used in order to find the winner tuple. A method using template matching and mathematical morphology is implemented following the proposed methodology. RESULTS A success is achieved if the distance between the automatically detected blob center and the actual structure center is less than or equal to one optic disc radius. The success rates obtained in the different public databases analyzed were: MESSIDOR (99.33%, 98.58%, 97.92%), DIARETDB1 (96.63%, 100%, 97.75%), DRIONS (100%, n/a, 100%) and ONHSD (100%, 98.85%, 97.70%) for optic disc (OD), macula (M) and vascular bundle (VB), respectively. Finally, the overall success rate obtained in this study for each structure was: 99.26% (OD), 98.69% (M) and 98.95% (VB). The average time of processing per image was 4.16 ± 0.72 s. CONCLUSIONS The main advantage of the use of inter-structure relational knowledge was the reduction of the number of false positives in the detection process. The implemented method is able to simultaneously detect four structures. It is fast, robust and its detection results are competitive in relation to other methods of the recent literature.

Localization and segmentation of the optic nerve head in eye fundus images using pyramid representation and genetic algorithms

This paper proposes an automatic method to locate and segment the optic nerve head (papilla) from eye fundus color photographic images. The method is inspired in the approach presented in [1]. Here, we use a Gaussian pyramid representation of the input image to obtain a subwindow centered at a point of the papillary area. Then, we apply a Laplacian pyramid to this image subwindow and we obtain a set of interest points (IPs) in two pyramid levels. Finally, a two-phase genetic algorithm is used in each pyramid level to find an ellipse containing the maximum number of IPs in an offset of its perimeter and, in this way, to achieve a progressive solution to the ONH contour. The method is tested in an eye fundus image database and, in relation to the method described in [1], the proposed method provides a slightly lower performance but it simplifies the methodology used to obtain the set of IPs and also reduces the computational cost of the whole process.

Segmentation of Moving Objects with Information Feedback Between Description Levels

In real sequences, one of the factors that most negatively affects the segmentation process result is the existence of scene noise. This impairs object segmentation which has to be corrected if we wish to have some minimum guarantees of success in the following tracking or classification stages. In this work we propose a generic knowledge-based model to improve the segmentation process. Specifically, the model uses a decomposition strategy in description levels to enable the feedback of information between adjacent levels. Finally, two case studies are proposed that instantiate the model proposed for detecting humans.