Influence of camera properties on image analysis in visual tunnel surveillance

Abstract

CCTV and image analysis systems for automatic incident detection are major tools in tunnel safety management. This paper presents results of a feasibility study called VITUS−1. Among others, one aim of VITUS-1 was to compare image analysis methods for robust object detection, tracking and incident detection. Image quality improvements were shown by using digital camera technology instead of state-of-the-art analogue CCTV. To improve image quality and robustness of existing image analysis methods, requirements on several camera properties were defined. We want to sensitise primarily tunnel operators and tunnel CCTV experts that commercial image analysis systems usually do not claim such requirements, e.g. requirements on resolution. This paper shows with a simple camera model that a robust detection of test objects is hardly feasible with typical tunnel CCTV (PAL resolution, 20 • viewing angle, 212m camera distance). As a consequence, the specifications of tunnel CCTV should be adjusted to the needs of image analysis systems. Simple experiments with real video data show the applicability of the camera model.

Non-Overlapping Multi-camera Detection and Tracking of Vehicles in Tunnel Surveillance

“…In order to evaluate the impact of our tracking method on the overall system performance, inter-camera tracking and multi-camera matching were evaluated for three scenarios: Pflugfelde: tracking by detection method using the nearest neighbour metric and a Kalman Filter, as described in [19]. ; SinglePart: detection and tracking strategies that we propose in this article, excluding interaction and error correction; Proposed: our proposed tracking methodology.…”

“…These are challenging tasks given the harsh illumination conditions usually found in tubular passages that are artificially illuminated. Therefore, despite continuous progress of computer vision algorithms for detection and tracking of vehicles, there is still a lack of accuracy in the performance of the techniques available in the state-of-the-art [1] when they are applied in tunnel surveillance [19]. This is mainly due to two reasons.…”

“…To the best of our knowledge, authors in [19] are the only ones to approach specifically the problem of visual surveillance in tunnel environments. They developed a system (VITUS) for detection and tracking of objects through several cameras.…”

Deep Learning-Based Consistent Object Detection in Distance on Highly Perspective Region

“…CCTVs in the tunnels are normally installed at lower positions compared to those on open roads, as illustrated in Figure 1. Therefore, these CCTV installations lead to severely distorted perspective footage, and this image distortion causes poor recognition of vehicles or humans as the relative distance between the CCTVs and the identified objects increases [13]. In particular, this mechanism has disadvantages in object detection (OD) performance using computer vision techniques [14].…”