Abstract

In this paper, we propose a background subtraction algorithm specific for depth videos from RGB-D cameras. Embedded in a people detection framework, it does not classify foreground / background at pixel level but provides useful information for the framework to remove noise. Noise is only removed when the framework has all the information from background subtraction, classification and object tracking. In our experiment, our background subtraction algorithm outperforms GMM, a popular background subtraction algorithm, in detecting people and removing noise.

We present a new evaluation methodology to better evaluate video processing performance. Recent evaluation methods [10], [9], [11] depend heavily on the benchmark dataset. The result may be different if we change the testing video sequences. The difference is mainly due to the video sequence content which usually includes many video processing problems (illumination changes, weak contrast etc.) at different difficulty levels. Hence it is difficult to extrapolate the evaluation result on new sequences.In this paper, we propose an evaluation methodology that help to reuse the evaluation result. We try to isolate each video processing problem and define quantitative measures to compute the difficulty level of a video relatively to the given problem. The maximum difficulty level of the videos at which the algorithm is performing good enough is defined as the upper bound of the algorithm capacity for handling the problem. To illustrate this methodology, we present metrics that evaluate the algorithm performance relatively to the problems of handling weakly contrasted objects and shadows.

This paper presents a controller for background subtraction algorithms to detect mobile objects in videos. The controller has two main tasks.The first task is to guide the background subtraction algorithm to update its background representation. To realize this task, the controller has to solve two important problems: removing ghosts (background regions misclassified as object of interest) and managing stationary objects. The controller detects ghosts based on object borders. To manage stationary objects, the controller cooperates with the tracking task to detect faster stationary objects without storing various background layers which are difficult to maintain.The second task is to initialize the parameter values of background subtraction algorithms to adapt to the current conditions of the scene. These parameter values enable the background subtraction algorithms to be as much sensitive as possible and to be consistent with the feedback of classification and tracking task.

In this paper, we propose a shadow removal algorithm for indoor scenes. This algorithm uses three types of constraints: chromaticity consistency, texture consistency and range of shadow intensity. The chromaticity consistency is verified in both HSV and RGB color spaces. The texture verification is based on the local coherency (over a pixel neighbourhood) of intensity reduction ratio between shadows and background. Finally, for the range of shadow intensity, we define a localized lower bound of the intensity reduction ratio so that dark mobile objects are not classified as shadows. Because the chromaticity constraint is only correct if the chromaticity of ambient light is the same as that of diffuse light, our algorithms only works in the indoor scenes.

We present a new method to detect abnormal gait based on the symmetry verification of the two-leg movement. Unlike other methods requiring special motion captors, the proposed method uses image processing techniques to correctly track leg movement. Our method first divides each leg into upper and lower parts using anatomical knowledge. Then each part is characterised by two straight lines approximating its two borders. Finally, leg movement is represented by the angle evolution of these lines. In this process, we propose a new line approximation algorithm which is robust to the outliers caused by incorrect separation of leg into upper / lower parts. In our experiment, the proposed method got very encouraging results. With 281 normal / abnormal gait videos of 9 people, this method achieved a classification accuracy of 91%.

Significant RF power deposition in the body causing local specific absorption rate (SAR) in the form of hotspots is an important safety concern at 3T (128 MHz) and, even more so, at 7T (298 MHz). In this work, we expand the proof-of-concept of artificial intelligence based real-time MRI safety prediction software (MRSaiFE) to 10 body models. We show that SAR patterns can be predicted with a mean squared error (MSE) of less than 1% and a structural similarity index of above 90% for 7T brain and above 85% for 3T body MRI.