The goal of this article is to compare some optimised implementations on current high performance platforms in order to highlight architectural trends in the field of embedded architectures and to get an estimation of what should be the components of a next generation vision system. We present some implementations of robust motion detection algorithms on three architectures: a general purpose RISC processor—the PowerPC G4—a parallel artificial retina dedicated to low level image processing—Pvlsar34—and the Associative Mesh, a specialized architecture based on associative net. To handle the different aspects and constraints of embedded systems, execution time and power consumption of these architectures are compared.

https://cyberleninka.org/article/n/359025.pdf

High performance motion detection: some trends toward new embedded architectures for vision systems

Surveillance systems depend greatly on the robustness and availability of the video streams. The cameras must deliver reliable streams from an angle corresponding to the correct viewpoint. In other words, the field of view and video quality must remain unchanged after the initial installation of a surveillance camera. The paper proposes an approach to detect changes (such as displacement, out of focus, obstruction) automatically in a difficult environment with illumination variations and dynamic background and foreground objects.

Automated camera dysfunctions detection

The context of this work is to develop, adapt and integrate augmented reality related tools to enhance the emotion involved in cultural performances. Part of the work was dedicated to augmenting a stage in a live performance, with dance as an application case. In this paper, we present a milestone of this work, an augmented dance show that brings together several tools and technologies that were developed over the project's lifetime. This is the result of mixing an artistic process with scientific research and development. This augmented show brings to stage issues from the research fields of Human-Machine Interaction (HMI) and Augmented Reality (AR). Virtual elements are added on stage (visual and audio) and the dancer is able to interact with them in real-time, using different interaction techniques. The originality of this work is threefold. Firstly, we propose a set of movement-based interaction techniques that can be used independently on stage or in another context. In this set, some techniques are direct, while others go through a high level of abstraction. Namely, we performed movement-based emotion recognition on the dancer, and used the recognized emotions to generate emotional music pieces and emotional poses for a humanoid robot. Secondly, those interaction techniques rely on various interconnected systems that can be reassembled. We hence propose an integrated, interactive system for augmenting a live performance, a context where system failure is not tolerated. The final system can be adapted following the artist's preferences. Finally, those systems were validated through an on field experiment - the show itself - after which we gathered and analyzed the feedback from both the audience and the choreographer.

/pdf/interactions-and-systems-for-augmenting-a-live-dance-533mrdgd2a.pdf

Interactions and systems for augmenting a live dance performance

We present in this paper a new system based on motion JPEG2000 intended for road surveillance application. The system uses a reference image and consists in 4 processing steps, namely initialization phase where the first reference image is built, reference estimation, motion segmentation (foreground extraction, ROI mask), and JPEG2000 coding. A first order recursive filter is used to build a reference image that corresponds to the background image. The obtained background is sent to the decoder once for all. The reference image at the coder side is estimated according to a Gaussian mixture model. The remote reference image is updated when specific conditions are met. The updating remote reference is triggered according to the states of mobile objects in the scene (no, few or lot of mobiles). The motion detection given by classical background subtraction technique is performed in order to extract a binary mask. The motion mask gives the region of interest of the system. The JPEG2000 image coded with a ROI option is sent towards the decoder. The decoder receives, decodes the image and builds the implicit binary ROI mask. Then, the decoder builds the displayed image using the reference image, the current image and the mask.

https://hal.archives-ouvertes.fr/hal-00199481/document

Dynamic Background Segmentation for Remote Reference Image Updating within Motion Detection JPEG2000

This paper presents various classifiers results from a system that can automatically recognize four different static human body postures in video sequences. The considered postures are standing, sitting, squatting, and lying. The three classifiers considered are a naive one and two based on the belief theory. The belief theory-based classifiers use either a classic or restricted plausibility criterion to make a decision after data fusion. The data come from the people 2D segmentation and from their face localization. Measurements consist in distances relative to a reference posture. The efficiency and the limits of the different classifiers on the recognition system are highlighted thanks to the analysis of a great number of results. This system allows real-time processing.

/pdf/belief-theory-based-classifiers-comparison-for-static-human-5dgtsu0kp3.pdf

Belief Theory-Based Classifiers Comparison for Static Human Body Postures Recognition in Video

This paper introduces a video object segmentation algorithm developed in the context of the European project Art.live where constraints on the quality of segmentation and the processing rate (at least 10 images/second) are required. In order to obtain a fine segmentation (no blocking effect, boundaries precision, temporal stability without flickering), the segmentation process is based on Markov random field (MRF) modelling which involves consecutive frame difference and a reference image in a unified way. Temporal changes of the luminance are predominant when the reference image is not yet available whereas the reference image prevails for low textured moving objects or for objects which stop moving for a while. The increased processing rate comes from the substitution of some Markovian iterations with morphological operations without loss of quality. Simulation results show the efficiency of the proposed method in term of accuracy and complexity (/spl sime/6 images/second for 352/spl times/288 pixels YUV images on a low-end processor).

A. Caplier

Papers

Robust fast extraction of video objects combining frame differences and adaptive reference image