Jean-Charles Quinton

Ant-Cams Network: a cooperative network model for silly cameras

This paper describes a new model of camera networks that can be used for environment monitoring and understanding. Such networks can be composed of both smart cameras, which benefit from high resolutions, powerful processing capabilities and strategic viewpoints on the environment, or with what we may call silly cameras, defined by much lower specifications. In this paper, we will demonstrate how our approach can reach efficient high-level understanding in spite of the limited information provided by each silly camera. We thus introduce the Ant-Cam model, inspired from the world of ants, which are able to solve complex problems by communicating despite their limited capabilities. The main idea is that data exchanged between the cameras is as important as the information extracted locally. Fully exploiting the cameras interactions without prior knowledge about the network configuration (e.g. camera location), the system is able to learn regularities and then infer from distributed sequences of events, passed between Ant-Cams. Once the system reaches a steady state, useful information can be inferred, such as the most used path or if space covering is sufficient.

Distributed coordination model for smart sensing applications

Distributed networks of smart sensors are nowadays representing the frontier of Machine-to-Machine (M2M) interoperability. In such a scenario several challenges must be addressed in order to create effective solutions. Coordination among nodes to satisfy monitoring purposes while addressing network constraints is considered of utmost importance. In this respect, the paper proposes a novel coordination model for self-organizing smart monitoring systems. The proposed algorithm is able to autonomously retrieve event correlations from the environment in order to coordinate the nodes. By relying on temporal and spatial correlations, the proposed system can be particularly suited for Smart Camera Network (SCN) deployments where multiple cameras monitor distributed targets. Along the algorithm definition, the paper presents a performance evaluation of the proposed approach through simulations, thus evaluating the robustness of the proposed model against message losses.

A Holistic Approach for Optimizing DSP Block Utilization of a CNN implementation on FPGA

Deep Neural Networks are becoming the de-facto standard models for image understanding, and more generally for computer vision tasks. As they involve highly parallelizable computations, Convolutional Neural Networks (CNNs) are well suited to current fine grain programmable logic devices. Thus, multiple CNN accelerators have been successfully implemented on Field-Programmable Gate Arrays (FPGAs). Unfortunately, FPGA resources such as logic elements or Digital Signal Processing (DSP) units remain limited. This work presents a holistic method relying on approximate computing and design space exploration to optimize the DSP block utilization of a CNN implementation on FPGA. This method was tested when implementing a reconfigurable Optical Character Recognition (OCR) convolutional neural network on an Altera Stratix V device and varying both data representation and CNN topology in order to find the best combination in terms of DSP block utilization and classification accuracy. This exploration generated dataflow architectures of 76 CNN topologies with 5 different fixed point representation. Most efficient implementation performs 883 classifications/sec at 256 × 256 resolution using 8 % of the available DSP blocks.

SAIL-MAP: Loop-closure detection using saliency-based features

Loop-closure detection, which is the ability to recognize a previously visited place, is of primary importance for robotic localization and navigation problems. We here introduce SAIL-MAP, a method for loop-closure detection based on vision only, applied to topological simultaneous localization and mapping (SLAM). Our method allows the matching of camera images using a novel saliency-based feature detector and descriptor. These features have been designed to benefit from the robustness to viewpoint change and image perturbations of bio-inspired saliency algorithms. Additionally, the same algorithm is used for the detector and descriptor. The results obtained on different large-scale data sets demonstrate the efficiency of the proposed solution for localization problems.

Ant-Cam network: Tracking across cameras using SVT parameters

Over the last years, the rapid growth of distributed smart cameras has triggered the search for new approaches of smartness of cameras to have better results. As communication among camera entities is becoming more and more complex and new ways of modeling communication have been proposed. These new ways have been taking inspiration from different fields such as socio-economic approach or game theory. Moreover, one of the major problems of the camera network is re-identification. However, in most cases, the interaction between cameras presupposes that the latter are able to perform perfect and unambiguous detections, thus limiting the decision tasks to the Markovian model. Within this paper, we present a new approach of interaction between cameras based on a non-Markovian model. To resolve this issue, we can exploit other types of information rather than visual information to improve re-identification. This information is Spatial, Visual and Temporal (SVT). Temporal information holds the time needed to go from one camera to another, while spatial information contains the path followed by the target which is a key point for the decision-making process. This offers the possibility for the network to learn regularities and then reach a steady state.

LobNet: A low specification camera network platform using Ant-Cam

Most of the existing platforms of smart camera networks rely on high-specification cameras with a high resolution visual sensor and high performed processors to provide high efficiency. However, for ethics and privacy concern, or to reduce computation requirement and power consumption, we can opt for low resolution. In this paper, we present a new platform LobNet for low-specification cameras Ant-Cam using a mouse sensor providing 30*30 pixels. An FPGA is used for processing which allows having completely distributed computing. Also, SmartMesh IP is utilized for communication. Low resolution can not provide much information and the silhouette observed by cameras may completely change depending on the coming point. However, focusing on this transformation between each pair of cameras can provide re-identification information.

Dual filtering in operational and joint spaces for reaching and grasping.

To study human movement generation, as well as to develop efficient control algorithms for humanoid or dexterous manipulation robots, overcoming the limits and drawbacks of inverse-kinematics-based methods is needed. Adequate methods must deal with high dimensionality, uncertainty, and must perform in real time (constraints shared by robots and humans). This paper introduces a Bayesian filtering method, hierarchically applied in the operational and joint spaces to break down the complexity of the problem. The method is validated in simulation on a robotic arm in a cluttered environment, with up to 51 degrees of freedom.