Patrik Andersson

Pedestrian Detection Using Augmented Training Data

Detecting pedestrians is a challenging and widely explored problem in computer vision. Many approaches rely on large quantities of manually labelled training data to learn a pedestrian classifier. To reduce the need for collecting and manually labelling real image training data, this paper investigates the possibility to use augmented images to train a pedestrian classifier. Augmented images are generated by rendering virtual pedestrians onto real image backgrounds. Classifiers learned from real or augmented training data are evaluated on real image test data from the widely used Daimler Mono Pedestrian benchmark data set. Results show that augmented training data generated from a single 200 frame image sequence reach 70 % average detection rate at one False Positives Per Image (FPPI), compared to 81 % for a classifier trained by a large-scale real data set. Results also show that complementing real training data with augmented data improves detection performance, compared to using real training data only.

Intensity Potential Approach for Modeling High-Frequency Sound Fields

This paper proposes the intensity potential approach for prediction of high-frequency sound power radiation. The approach is based on the Helmholtz decomposition of the vector field of time-averaged sound intensity into its irrotational and rotational components. The local power balance in a lossless medium is expressed in terms of the irrotational component only, and results in the Poisson equation for a scalar intensity potential of this component only. The approach gives an exact expression for the sound power through any closed surface in terms of the irrotational component, provided that the boundary conditions are correct. The approach is evaluated by exploring the two intensity components in three canonical examples, and by comparison to measured data with special focus on directivity aspects. It is concluded that the intensity potential approach is relevant, in particular for high-frequency sound fields from multiple sources that are uncorrelated and broadbanded. However, the intensity is generally overestimated in the shadow zones and underestimated in the directly exposed regions. Further, peaks in narrow frequency bands associated with interference of waves are ignored.

Boundary Element Method for Intensity Potential Approach : Predicting the Radiated Sound Power from Partially Enclosed Noise Sources

This paper proposes the boundary element method for the intensity potential for prediction of high-frequency sound power flow through partial enclosures. The intensity potential approach is based on the local power balance in a lossless medium and the Helmholtz decomposition of the vector field of time-averaged sound intensity. The result is a Poisson equation for a scalar intensity potential. The intensity potential formulation and the boundary element method are both suitable for exterior problems. The governing equations of the intensity potential and the boundary element method for solving this problem are presented. Results from the proposed method are compared with experimental results, for the case of radiated sound power in one-third-octave bands from sources in a partial enclosure. The results show that the method is applicable for estimation of global radiated sound power in one-third-octave bands in the high-frequency range.

An impedance matching technique for active‐passive vibration control

Impedance matching techniques have shown potential for active vibration control of structures in bending. Such structures are commonly described by Euler-Bernoulli theory. Previous studies concerning impedance matching of these structures have only considered scalar quantities. However, for an Euler-Bernoulli beam four field variables are involved which implies that a scalar impedance is insucient. The purpose of this study is therefore to expand the technique to include full 2x2 matrices. This is achieved by first deriving the reflection matrix as a function of the characteristic impedance matrices of an Euler-Bernoulli beam and an arbitrary termination impedance. An active impedance load is then introduced in order to manipulate the reflection matrix. A theoretical example is given where the approach is utilized to match the junction between an Euler-Bernoulli beam and a sandwich composite. This proposed active-passive damping configuration employs active control to enclose all incident power in the sandwich composite. Results show that the active impedance load is responsible for the main part of the power absorption over a broad frequency range.