Was it a fall or were they pushed?
The NFI and TU Delft are working on a model for complex fall cases. Suppose that a deceased person is lying at the bottom of the stairs in a house: did that person take a fall, were they pushed or killed beforehand and placed there to make it look like an accident? These are often highly complex cases for law enforcement agencies, which may stall or be approached incorrectly. In order to change this, the Netherlands Forensic Institute (NFI) has been working on a computer model with TU Delft that is able to simulate human falls. In future, the model should be able to assist the police and the Public Prosecution Service in determining the most likely scenario of what happened.
Test falls with subjects and foam block pit
The way someone is lying, i.e. their position, will often already give the police an idea of what may have happened. For example, if the person is lying in a conspicuous position, investigators at the scene will attempt to identify what may have happened and whether the traces match a particular scenario.
‘However, occasionally what you may see at first glance will differ from what actually happened,’ says Jan Peter van Zandwijk, a forensic scientist and the project leader, ‘ which is why a computer model could assist the investigation.’
Model simulation
That is why the NFI launched the Icarus project two years ago. The researchers arrived at an existing commercial computer model that was developed for the automotive industry. Van Zandwijk and Kim Hutchinson, a BioMechanical Engineering student at TU Delft, have been testing this model over the past two years with the aim of examining whether the model could be used for forensic investigation.
‘You might compare it to a “crash test dummy”, which is a dummy that they use to simulate car accidents’, Hutchinson explains. ‘Only our model is a virtual dummy – a human dummy inside a computer. The model calculates how the human dummy were to fall if you were to place it in a certain position and were to push it with a certain amount of force, based on the laws of physics.’
The dummy can be placed in a variety of starting positions and a fall can then be simulated to determine where the dummy ends up. The researchers can use the model to simulate a fall, but can equally simulate a movement resulting in someone having been pushed, for example.
A soft landing
In order to gain a better understanding of the falling behaviour of real people and to be able to compare it with the way the model falls, the researchers conducted fall tests with test subjects. The fall tests involved subjects jumping, falling or being pushed from a height of 2.5 metres.
Naturally, the test subjects had a soft landing, ending up in a large pit filled with foam blocks with a trampoline as the undersurface.
‘We then began measuring the test subjects in as much detail as possible. How do all the joints move in relation to each other? What force does the body exert on the ground? To allow for these measurements, test subjects wore a special suit during their fall or jump containing sensors that recorded their movements’, says Hutchinson.
‘We also measured the pushing force using an instrument that I designed especially for that purpose, to gain as much knowledge as possible about the force impacting the test subjects.’
Recreating the crime scene
The researchers then entered the same starting position and forces into the computer model to see if the virtual dummy would move in the same way as the test subjects, examining the behaviour exhibited during a fall as well as what that type of fall looks like. The researchers can also assign the virtual dummy the same sex, weight, and height as the ‘victim’.
‘The model already included those options’, says van Zandwijk. ‘On top of that, we can also build a virtual staircase within the model, for example, in order to simulate a crime scene. Fortunately, the interaction between the virtual dummy and those elements already worked well beforehand’, says Van Zandwijk.
Various scenarios
The researchers’ ultimate goal is to compare the traces at the crime scene with the model simulation in order to arrive at a probability statement regarding a particular scenario.
Are the traces more probable under the scenario that someone fell, jumped or do the traces fit more closely with the scenario that someone was pushed down the stairs. Given that many of the circumstances at the crime scene are often unknown, the researchers conduct multiple simulations of falling, jumping and pushing movements using the virtual human dummy, with the starting position or pushing force, for example, varying slightly each time.
It can then be established in which of these simulations the dummy best fits and whether it corresponds to the position of the victim at the real crime scene. ‘This is how we try to distinguish between the various scenarios’, Van Zandwijk explains.
Follow-up research
Nevertheless, following an analysis of the data, it has been concluded that the computer model does not yet work for this type of application as the researchers had envisaged.
‘The virtual dummy falls very passively, which still does not really emulate how real people fall: they tend not to fall down like a sack of potatoes when pushed. People are frightened of falling down and react by breaking their fall. They don’t fall upside down with their head angled downward', Hutchinson explains.
This is something the computer model still does. ‘We will continue to focus on that moving forward’, says Van Zandwijk, ‘and we intend to carry on in October with a new researcher from TU Delft’, as Hutchinson’s graduation project has now been completed.
The forensic scientist says that they intend to identify the problem in the model through follow-up research. ‘Hopefully this is something that can be solved easily. We can then align the behaviour of the virtual dummy more closely with that of humans.’
He also hopes to gain more insight into what type of movements work best on the model and therefore for what types of falls the model can be used reliably in future. ‘These types of complex falls occur regularly, which means there is definitely a need for a supporting computer model.’