Multi-scale Covariance Descriptor Research Articles

A Trainable System for Underwater Pipe Detection

Underwater image processing is widely increased over the last decade. It is a fundamental process for a most part of underwater research applications, because of the need of data acquisition. In this paper we will propose a novel approach of pipe detection in submarine environment. The system draws much of its power from a representation that describes an object class taking into account structure and content features which are computed through the multi-scale covariance descriptor. This approach describes an object detection model by training a support vector machine classifier using a large set of positive and negative samples. We present result on pipe detection using Maris dataset. Moreover, we show how the representation affects detection performance by considering mono-scale representation using Covariance descriptor.

An improvement of multi-scale covariance descriptor for embedded system

Video surveillance has been a major area of focus for researchers and engineers. Actually, video surveillance includes several useful and complex tasks such as tracking, human detection, re-identification and recognition. Multi-scale covariance (MSCOV) descriptor has recently grown in interest due to its good performances for person detection, re-identification and matching. Unfortunately, its original version requires heavy computations, and it is difficult to be executed in real time on embedded systems. This paper presents two aspects of improvement to adapt the MSCOV descriptor for embedded systems. First, the local binary pattern (LBP) features are introduced and a trade-off between accuracy and processing cost is used to define the best features combination. Second, parallel implementation and embedded co-processor are exploited to accelerate processing time on multi-core CPU architectures. Both optimizations are implemented and evaluated for executing a complete application of person re-identification systems. The software implementation is performed using the VIPeR dataset. Using LBP, 21.57% processing speed-up and 50% less memory requirements for the descriptor computation are achieved without any accuracy performance degradation. We also prototype the proposed design using Zynq platform based on ARM Cortex-A9. The results demonstrate the effectiveness of the parallelization and conduct more than 11 times processing speed-up against the original algorithm.

A 3D Scene Registration Method via Covariance Descriptors and an Evolutionary Stable Strategy Game Theory Solver

In this paper we provide an integrated approach for matching patterns in scenes combining 3D and visual information. For local definition of points we propose a descriptor based on the notion of covariance of features for fusion of shape and color information of 3D surfaces, so-called multi-scale covariance descriptor (MCOV). The intrinsic properties of this descriptor are many: it is invariant to spatial rigid transformations, and robust to noise and resolution changes; it can also be used for characteristic point detection; and lies on top of a manifold topology which allows the use of analytical metric properties. This descriptor is complemented with a game theoretic approach for solving the matching correspondences under global geometric constraints. This layer offers a comprehensive understanding of the scene and avoids possible mismatches due to repeated areas or symmetries--which would be impossibly identified by the detector solely at a local level. Our solution is able to accurately match different views of a scene even under spatial transformations, high noise levels and with small overlap between views, outperforming state-of-the-art approaches. Results are validated by comparing MCOV against other state-of-the-art 3D point descriptor methods, and matching complex 3D and color scenes under several challenging conditions.

A fast multi-scale covariance descriptor for object re-identification

In many surveillance systems, there is a need to determine if a given object (person, group of persons, vehicle, …) has already been observed over a network of cameras. It is the object re-identification problem. Solving this problem involves matching observation of objects across disjoint camera views. Uncalibrated fixed or mobile cameras with non-overlapping field of view generate uncontrolled variation in view point, background and lighting. In such situations, a robust and invariant image description is required. A multi-scale covariance image descriptor and a quadtree based scheme are proposed to describe any object of interest. We describe a fast method for computation of multi-scale covariance descriptor. The descriptor is evaluated in person re-identification application using the VIPeR dataset. We show that the proposed multi-scale approach outperforms existing mono-scale image description methods.