Object detection in road scenes is an unavoidable task to develop autonomous vehicles and driving assistance systems. Deep neural networks have shown great performances using conventional imaging in ideal cases but they fail to properly detect objects in case of unstable scenes such as high reflections, occluded objects or small objects. Next to that, Polarized imaging, characterizing the light wave, can describe an object not only by its shape or color but also by its reflection properties. That feature is a reliable indicator of the physical nature of the object even under poor illumination or strong reflections. In this paper, we show how polarimetric images, combined with deep neural networks, contribute to enhance object detection in road scenes. Experimental results illustrate the effectiveness of the proposed framework at the end of this paper.
A nonparametric method to define a pixel neighborhood in catadioptric images is presented. The method is based on an accurate modeling of the mirror shape by mean of polarization imaging. Unlike most processing methods existing in the literature, this method is nonparametric and enables us to respect the catadioptric image’s anamorphosis. The neighborhood is directly derived from the two polarization parameters: the angle and the degree of polarization. Regardless of the shape of the catadioptric sensor’s mirror (including noncentral configurations), image processing techniques such as image derivation, edge detection, interest point detection, as well as image matching, can be efficiently performed.
The ability of a robot to localise itself and simultaneously build a map of its environment (Simultaneous Localisation and Mapping or SLAM) is a fundamental characteristic required for autonomous operation of the robot. Vision Sensors are very attractive for application in SLAM because of their rich sensory output and cost effectiveness. Different issues are involved in the problem of vision based SLAM and many different approaches exist in order to solve these issues. This paper gives a classification of state-of-the-art vision based SLAM techniques in terms of (i) imaging systems used for performing SLAM which include single cameras, stereo pairs, multiple camera rigs and catadioptric sensors, (ii) features extracted from the environment in order to perform SLAM which include point features and line/edge features, (iii) initialisation of landmarks which can either be delayed or undelayed, (iv) SLAM techniques used which include Extended Kalman Filtering, Particle Filtering, biologically inspired techniques like RatSLAM, and other techniques like Local Bundle Adjustment, and (v) use of wheel odometry information. The paper also presents the implementation and analysis of stereo pair based EKF SLAM for synthetic data. Results prove the technique to work successfully in the presence of considerable amounts of sensor noise. We believe that state of the art presented in the paper can serve as a basis for future research in the area of vision based SLAM. It will permit further research in the area to be carried out in an efficient and application specific way.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.