Search

Localization-loss and classification-loss are optimized at the same time to train the one-stage object detectors. Because of the large number of anchors, the severe foreground–background class disproportion causes significant classification-loss. This paper discusses using a ranking module instead of the classification module to mitigate this difficulty and also Average-Precision loss (AP-loss) is utilized on the ranking module. An optimization algorithm is used to make the AP-loss as effective as possible. Optimization algorithm blends the error-driven updating method of perceptron learning and the deep network backpropagation technique. This optimization algorithm handles the foreground–background class disproportion issues. One-stage detector with AP-loss and backbone with ResNet-152 attains improvement in the detection performance compared to the classification-losses-based detectors. [ABSTRACT FROM AUTHOR]

In this paper, we propose a watermarking scheme for non-uniform B-spline (NUBS) surface. Firstly, we first do sampling on a NUBS surface and get the sample points, then the watermark is embedded into the DCT coefficients of the sample points and the watermarked sample points are transformed back, finally the watermarked surface is reconstructed from watermarked sample points using global interpolation. A sign correlation detector is used to test for the presence of the watermark, and the original surface is not required at this stage. Experimental results show that our algorithm can preserve the shape of the original surface within a specified error, and that it is robust against attacks including knot insertion, order elevation, addition of white noise, rotation, scaling, translation and further watermarking. [ABSTRACT FROM AUTHOR]

Video surveillance cameras are ubiquitous this decade. With the popularization of sports, costly courts are built widespread. To protect the sport ground against from damage, some activities, such as biking and in-line skating, are prohibited in the court. Traditional video surveillance systems can not prevent these activities in time. In this paper, we propose a video-based approach for detecting prohibited activities that can damage a court. The approach involves prohibited activity analysis and prohibited activity detection. The first stage generates the leg-angle curves of a prohibited activity. A background subtraction procedure is applied to extract moving objects; moving objects are then normalized to minimize influences of various scaling conditions. Leg-angle curves of prohibited activities are generated by computing leg angles. The second stage detects specific prohibited activities by analyzing leg-angle curves obtained from the input video sequences. Our proposed method can detect the prohibited activities in a court, thus preventing such activities in time. [ABSTRACT FROM AUTHOR]

Edge detection concerns the localization of significant variations of the grey level image. Detection of edges in an image is very important step for a complete image understanding system. This paper proposes a new approach to edge detection which adopts fuzzy reasoning to detect edges and mathematical morphology for edge thinning. The results achieved by this algorithm are comparable to the Canny approach. In Canny edge detector we may require many runs using different combinations of the three parameters (two threshold and one sigma values) whereas in the proposed technique only one parameter needs to be set by the user coarsely to get the same results, also the computation load of Canny is higher than the proposed approach. [ABSTRACT FROM AUTHOR]

Artificial Color is the use of technology to emulate the basic means by which animals acquire and use spectral information. It begins with sensing of the scene using multiple broad and overlapping spectral sensitivity functions analogous with cone cells in the human visual system. In this paper we develop an algorithm to obtain a set of near-optimal application-specific spectral sensitivity functions for Artificial Color systems. The user-designed sensitivity functions result in concurrent good intra-class packing and interclass integrity in the color space for reflectance classes with known statistics. This method leads to improved color discrimination and results in more robust spectral classifiers. [ABSTRACT FROM AUTHOR]

In this paper, a complete Smart Space architecture and related system prototype are presented. The system is able to analyze situations of interest in a given environment and to produce related contextual information. Experimental results show that video information plays a major role for what concerns both situation perception and personalized contex-aware communications. For this reason, the poropsed multisensor system automatically extracts information from multiple cameras as well as diverse sensors describing environment status. This information is then used to trigger personalized and context-aware video messages adaptively sent to users. A rule-based module is encharged to customize video messages in relation to the user profile, contextual situation and users's terminal. The systems outputs graphically generated video messages consisting of an animated avatar (i.e. Virtual Character) closing the loop on users. Proposed results validate the conceptual schema behind the architecture and the successful adaption to the analysis of different situations. [ABSTRACT FROM AUTHOR]

This article deals with an estimation method of 3D structure and motion. The object is described by line segments and points assuming that it can be described by a polygonal model. It models the problem which is solved by a nonlinear estimation method: The Particle Filter. This method can account for nonlinear models and non-Gaussian statistics without any linearization stage like the Extended Kalman Filter (EKF), for example. The increase in accuracy is shown on a vision system composed by sensors delivering range and intensity/reflectance image sequences. Finally, the solution is compared with a commonly used state estimation method (EKF). [ABSTRACT FROM AUTHOR]