Abstract
44
can access the data and programs, hacking the computers on board for criminal purpose for example: ● ● ● ●
To cheat insurance company To delete digital evidence (speed, time, date and time) To cause a car crash To misuse the vehicle for a terrorism attack
doi:10.1016/j.scijus.2009.11.065
20 years of experience in forensic cockpit voice recorder analysis: Potentiality, results, and recommendations F. Marescal Forensic Laboratory, French Gendarmerie, Rosny-sous-Bois, France In the field of air crash, the analysis of CVR (Cockpit Voice Recorder) is sometimes the only way to explain the circumstances of the crash, either to improve safety for future flights or to determine the judicial responsibilities. Our forensic laboratory (IRCGN) is involved in the judicial process. IRCGN works on the analysis of the flight recorders of any type of aircraft: civilian or military, private as soon as a decision of justice is committed. The first recorder analysed came from the HABSHEIM crash (east of France) from an Air France Airbus in 1989 during an air show. Then a dozen of cases have been investigated like the CONCORDE crash in July 2000. Standards set by the Convention on International Civil Aviation Organisation and the well-known ‘Annex 13’ are used for technical investigations. French ‘Bureau Enquête Analyse’, a specialised independent civilian aviation attached to the ministry of transport is in charge of the inquiries and analysis in aircraft accident matter to prevent new occurrences whereas our laboratory takes part in the judicial expertise to emphasise the penal responsibilities. After data extraction from the CVR, the laboratory picks out as much information as necessary to determine the event course: – Conversations of the different audio tracks are transcribed in order to understand the exchanges between pilots, cabin crew, and the control tower. Most often, ambient and interference noise has to be reduced – In case of suspicion of manipulation because the CVR was removed from the hand of justice while tape authentication could be performed – Specific noise within the cockpit (switch, lever, alarm, in board electric frequency) could be identified from the one perceived from the cockpit like an explosion or the regime of an engine – Robust speaker identification could be helpful to determine which of the pilots gives order – Emotional state assessment from the voice could be useful to understand their stress level and then their behaviour During this process, the expert is really an adviser for the magistrate who isn't aware of the kind of results that could be obtained. All these operations have to be conducted quickly in close relation with an expert pilot qualified in the type of aircraft. Then, more accurate analysis can be proceeded according to the inquiry. We propose a presentation that will concentrate on essential issues, audio examples, and on a few lessons learnt. doi:10.1016/j.scijus.2009.11.066
Crime scene or crash scene 3-D modelling L. Chartier, C. Lambert, J. Carlier Gendarmerie Nationale, IRCGN, Rosny-sous-Bois, France Since September 2007, the Forensic Science Institute of the French Gendarmerie (IRCGN) has been deploying a 3-D laser/scanner to fulfil its modelling capabilities with respect to crime/crash scenes. Its technical features make it a fast and accurate tool, bringing 3-step forensic
solutions. In a first step, the laser acquisition will ‘freeze’ the crime scene and its components. In a second step, the resulting virtual crime/crash scene model is used as a high performance tool, enabling miscellaneous examinations such as measurements, ballistic reconstructions etc. In a third step, the ‘signal, image and voice’ department has been developing a capability in the field of 3-D animation, in order to generate video clips to illustrate crime/crash sites or scenarios. This presentation describes how our department, in particular its Measurements and 3-D Modelling unit, is the driving force behind this novel approach, and how it associates many actors in their respective fields of expertise such as accident reconstructions, ballistics, blood pattern analysis, explosives and bomb attacks. Whenever measurements have to be performed on a crime/crash scene, the laser/scanner enables us to acquire the latter with a precision down to a millimetre. By multiplying station points, a large scene may thus be processed in a reasonable time. Results may be burnt onto a DVD and provided to investigators, CSIs, or magistrates. It comes with a free software, to view the 3-D model, move inside, and make measurements. 3-D modelling is perfectly suited to CSIs or ballistics experts to reconstruct trajectories or establish scenarios. In a case of blood pattern analysis, a module within the apparatus can compute and represent blood spatter trajectories to work out their point of origin in space. This may be carried out within minutes by overlapping high definition photographs to a model. Large scenes comprising an important number of physical evidences may also be managed. 3-D model processing can generate level surfaces, maps, 2-D/3-D maps, by superimposing in-flight photographs as well as 3-D animations. With two years experience in the field of 3-D scene processing, this presentation exposes accurate and reliable results in several cases. This presentation describes too how our unit carries on to enhance its application methods and subsequent processing from 3-D models in miscellaneous forensic topics.
doi:10.1016/j.scijus.2009.11.067
Linking cameras to images and videostreams with pixel response non-uniformity Z. Geradts, FIDIS NFI, Digital Technology and Biometrics, The Hague, Netherlands Efforts have been made within the European Project FIDIS (www.fidis. net) to examine and validate the methods for camera identification based on Pixel Response Non-Uniformity (PRNU). In this presentation, the current state of the art is given, with experiments with ten cameras of ten different makes and models, to validate how unique the patterns are within the groups. In the experiments, cameras such as phonecams were also used. These kinds of cameras use much JPEG-compression, and a filter has been used to filter out the JPEG-distortion. The first test with Youtube seemed to be difficult, since we could not easily distinguish the cameras from images that we have uploaded. We have implemented other methods with wavelet-filtering by Lukas et al. [1] which worked better in this test. The algorithms have been implemented in Java, and are open source at www.sourceforge.net, with the name ‘nfi prnu compare’. Also the database of reference cameras is available for download, in order to verify the results of this research. In future work, we will try to link cameras in child pornography databases from the police, and try to find images which have been made with the same camera based on the PRNU determination. Reference Lukáš, J., Fridrich, J., Goljan, M., June 2006. Digital camera identification from sensor noise. IEEE Transactions on Information Security and Forensics 1 (2), 205–214.
doi:10.1016/j.scijus.2009.11.068