IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 49, NO. 5, SEPTEMBER 2000 2013 A Distributed Surveillance System for Detection of Abandoned Objects in Unmanned Railway Environments Claudio Sacchi and Carlo S. Regazzoni, Senior Member, IEEE Abstract—In this paper, a distributed video-surveillance system for the detection of dangerous situations related to the presence of abandoned objects in the waiting rooms of unattended railway stations is presented. The image sequences, acquired with a monochromatic camera placed in each guarded room, are pro- cessed by a local PC-based image-processing system, devoted to detecting the presence of abandoned objects. When an abandoned object is recognized, an alarm issue is transmitted to a remote control center, located few miles far from the guarded stations. A multimedia communication system based on direct sequence code-division multiple-access (DS/CDMA) techniques aims at ensuring secure and noise-robust wireless transmission links between the guarded stations and the remote control center, where the processing results are displayed to the human operator. Re- sults concern: 1) the performances of each local image processing system in terms of false-alarm and misdetection probabilities, and 2) the performances of the CDMA multimedia transmission system in terms of bit error rates (BERs) and quality of service (QoS). Index Terms—Image processing, multimedia communication, rail transportation, site security monitoring, surveillance. I. INTRODUCTION T HE increasing request for security and efficiency in the field of public transportation systems, for both people and goods, has resulted in a corresponding increasing interest in the use of the most advanced video-based surveillance tech- niques in order to provide an automatic continuous monitoring of roads, railways, vehicles, and land transport infrastructures (e.g., railway stations, highway toll-gates, etc.). The main ob- jectives of a surveillance system in transport environments con- cern the detection and the prevention of dangerous situations, e.g., vehicle accidents, run-over pedestrians, people falling over railway tracks, cars that stopped at unattended level crossings, etc., and the management of the vehicular traffic, in order to op- timize the flow on roads and highways. Several applications of image processing and advanced data transmission techniques to the surveillance of transport environments have been presented in the literature [1]–[10]. Manuscript received January 4, 1999; revised February 15, 2000. This work was supported in part by the National Research Council (CNR) of Italy within the framework of the “National Targeted Project Transport 2” (PFT2) Research Programme, and by the Italian Ministery of University and Scientific Research (MURST) under the National Interest Scientific Research Programme. The authors are with the Department of Biophysical and Electronic Engineering (DIBE), University of Genoa, 16145, Genova, Italy (e-mail: carlo@dibe.unige.it). Publisher Item Identifier S 0018-9545(00)07906-8. Concerning road transport, the AUTOSCOPE system, devel- oped in the USA in the mid-1980s [1], is one of the best-known examples of video-based highway traffic monitoring systems. Image sequences acquired with a camera are processed by a mi- croprocessor system that detects in real time the presence or the passing of a vehicle in the camera field of view. Another system for real-time accident prevention and traffic monitoring has been developed in Europe and is described in [2]. The system, known as TRISTAR, processes images coming out from cameras placed near highway lanes and produces alarm signals when a poten- tially dangerous situation (e.g., accident risk) is detected. A fur- ther example of a video-based system for traffic monitoring and management is presented in [3]. The described system can be addressed to obtain a visual tracking modality (i.e., vehicle tracking and pedestrian tracking) for a traffic advisory system. In [3] it is shown that the exploitation of advanced image-pro- cessing techniques for moving-object detection and tracking can be a valid support to increase the margin of safety in a large va- riety of common traffic situations. A quite futuristic, though very interesting, research field is that of the video-based control procedures for computer-driven unmanned vehicles. In [4], a video-camera based method for de- termining the location and the rotation of autonomous vehicles is proposed. In [5] the development of a portable hardware/soft- ware neural-network module for autonomous vehicle following is described. An autonomous vehicle following is defined as a vehicle controlling its own steering and speed by following a lead vehicle [5]. A neural-network approach is exploited to determine the nonlinear relation between the observed range and heading angle and the controllable steering-wheel angle and speed. The data on the range and the heading angle are acquired by a stereo-vision system, and a neural-network-based image- processing system generates the driving command as its own issues. The synergies between vehicle recognition and tracking processes for autonomous vehicle driving are studied in [6]. Ob- ject recognition is performed in order to focus attention on inter- esting parts of a guarded scene and to assign symbolic meanings to them. Tracking is used to maintain a correspondence between the objects identified at successive recognition instants. In railway environments, traffic and car safety management tasks, such as: headway between trains, speed regulation, and collision avoidance, are generally implemented by railway signaling systems, using secure and noise-robust digital radio transmission techniques. For this reason, video-surveillance applications in railway transport essentially aim at meeting the request for a protection against accidental or intentional 0018–9545/00$10.00 © 2000 IEEE