Search

This paper proposes a generic methodology for the semi-automatic generation of reliable position annotations for evaluating multi-camera people-trackers on large video data sets. Most of the annotation data are automatically computed, by estimating a consensus tracking result from multiple existing trackers and people detectors and classifying it as either reliable or not. A small subset of the data, composed of tracks with insufficient reliability, is verified by a human using a simple binary decision task, a process faster than marking the correct person position. The proposed framework is generic and can handle additional trackers. We present results on a data set of $\sim 6$ h captured by 4 cameras, featuring a person in a holiday flat, performing activities such as walking, cooking, eating, cleaning, and watching TV. When aiming for a tracking accuracy of 60 cm, 80% of all video frames are automatically annotated. The annotations for the remaining 20% of the frames were added after human verification of an automatically selected subset of data. This involved $\sim 2.4$ h of manual labor. According to a subsequent comprehensive visual inspection to judge the annotation procedure, we found 99% of the automatically annotated frames to be correct. We provide guidelines on how to apply the proposed methodology to new data sets. We also provide an exploratory study for the multi-target case, applied on the existing and new benchmark video sequences.

The problem of foreground detection in real-time video surveillance applications is addressed. Proposes is a framework, which is computationally cheap and has low memory requirements. It combines two simple processing blocks, both of which are essentially background subtraction algorithms. The main novelty of the approach is a combination of an autoregressive moving average filter with two background models having different adaptation speeds. The first model, having a lower adaptation speed, models long-term background and detects foreground objects by finding areas in the current frame which significantly differ from the proposed background model. The second model, with a higher adaptation speed, models the short-term background and is responsible for finding regions in the scene with a high foreground object activity. The final foreground detection is built by combining the outputs from these building blocks. The foreground obtained by the long-term modelling block is verified by the output of the short-term modelling block, i.e. only the objects exhibiting significant motion are detected as real foreground objects. The proposed method results in a very good foreground detection performance at a low computational cost.

In this demo we present a people tracker in indoor environments. The tracker executes in a network of smart cameras with overlapping views. Special attention is given to real-time processing by distribution of tasks between the cameras and the fusion server. Each camera performs tasks of processing the images and tracking of people in the image plane. Instead of camera images, only metadata (a bounding box per person) are sent from each camera to the fusion server. The metadata are used on the server side to estimate the position of each person in real-world coordinates. Although the tracker is designed to suit any indoor environment, in this demo the tracker's performance is presented in a meeting scenario, where occlusions of people by other people and/or furniture are significant and occur frequently. Multiple cameras insure views from multiple angles, which keeps tracking accurate even in cases of severe occlusions in some of the views.

A novel approach to automatically detect vehicles in road tunnels is presented in this paper. Non-uniform and poor illumination conditions prevail in road tunnels making difficult to achieve robust vehicle detection. In order to cope with the illumination issues, we propose a local higher-order statistic filter to make the vehicle detection invariant to illumination changes, whereas a morphological-based background subtraction is used to generate a convex hull segmentation of the vehicles. An evaluation test comparing our approach with a benchmark object detector shows that our approach outperforms in terms of false detection rate and overlap area detection.

Since the end-user of video-based systems is often a human observer, prediction of user-perceived video quality (PVQ) is an important task for increasing the user satisfaction. Despite the large variety of objective video quality measures (VQMs), their lack of generalizability remains a problem. This is mainly due to the strong dependency between PVQ and video content. Although this problem is well known, few existing VQMs directly account for the influence of video content on PVQ. Recently, we proposed a method to predict PVQ by introducing relevant video content features in the computation of video distortion measures. The method is based on analyzing the level of spatiotemporal activity in the video and using those as parameters of the anthropomorphic video distortion models. We focus on the experimental evaluation of the proposed methodology based on a total of five public databases, four different objective VQMs, and 105 content related indexes. Additionally, relying on the proposed method, we introduce an approach for selecting the levels of video distortions for the purpose of subjective quality assessment studies. Our results suggest that when adequately combined with content related indexes, even very simple distortion measures (e.g., peak signal to noise ratio) are able to achieve high performance, i.e., high correlation between the VQM and the PVQ. In particular, we have found that by incorporating video content features, it is possible to increase the performance of the VQM by up to 20% relative to its noncontent-aware baseline.