A key method for the study of the perception and control of complex body movements is motion capture, a method for the highly accurate recording of complex human body movements. In order to reconstruct the movement trajectories of the body's joints, a large number of reflecting markers are attached to characteristic points (typically around 40) of the body surface.
Optical motion capture systems record the three-dimensional positions of the markers using multiple cameras, which typically work in the infrared and which also contain LEDs that emit infrared light in order to illuminate the markers. By applying appropriate algorithms for image processing the makers can be efficiently segmented from the image background. The three-dimensional marker positions are reconstructed from the two-dimensional camera images, filling in missing parts of trajectories by interpolation. Multiple groups in the CIN use VICON motion capture systems with sampling rates between 120 and 500 Hz and between 6 and 16 cameras. Movements are recorded as basis for the analysis of the motor behaviour, e.g. for the analysis of patients with movement disorders or to identify characteristic parameters of body movements that carry information, e.g. about emotion or other movement styles.
The same technology is applied as well in computer animation, e.g. for the creation of Hollywood movies such as Lord of the Rings. As a research tool motion capture is used by several groups within the CIN to generate visual stimuli to study perception of body motion and the simulation of virtual characters in virtual reality applications. The same technology is also useful for technical applications, such as character animation in computer graphics and the synthesis of human-like movements for robotics systems. Advanced work tries to read out the position data in real-time and uses it for the online animation of virtual characters, for example to study how executed movements influence the perception of body motion. In addition, several groups work on the extension of this technology to animals, for example to record arm and hand movements of nonhuman primates.
Research in computational vision and robotics has developed technical solutions of a number of relevant problems, e.g. in object recognition or imitation learning of movements. However, the high degree of flexibility and adaptability of visual information processing in biological systems has not yet been reached by any existing technical solution. Also technical systems have to be trained in a cumbersome procedure with many examples before they are able perform the required task. For instances machines that perform identification of faces, handwriting or other complex visual objects have to be confronted with millions of such object before they can delineate them in a trustworthy way.
Groups at the CIN work on transferring principles of information processing found in real brains to improve computer vision applications. Fundamental biological concepts, such as hierarchies of pattern detectors, learning of features at different levels of complexity or the implementation of bottom-up vs. top-down effects, are combined with modern approaches from machine learning, such as kernel machines or Bayesian inference methods to transfer these brain-like information processing to technical solutions. The prospect of this work is to introduce high flexibility and the possibility to train such systems with small amounts of data.
Neuroprosthetics is an exciting new field of interdisciplinary research involving medicine, neurophysiology, physics, engineering and information technology that has the prospect to lead to more potent and rational neurological therapies. Based on knowledge how neuronal networks represent information provided by basic research in systems neuroscience, it tries to ameliorate the deficits resulting from loss of neurons in sensory or central nervous structures by replacing them with technical systems interfacing with the brain.
One prominent example is the Cochlea Implant, a device that uses the tonotopy of the inner ear (‘cochlea’) to imprint sound information into the brain of deaf patients. Important insights have been gained by Helmholtz in the nineteenth century and von Bekesy in the twentieth amongst many others. The development was based on physical analysis and critical knowledge about function of the mechanical and neural auditory system gleaned from experiments in guinea pigs and cats.
The cochlea is a very complex and delicate structure, embedded in bone that contains three coiled, fluid-filled ducts delimited by membranes and beset by mechanoreceptors the so called hair cells. These specialized neurons convert sound-induced vibrations into electrical neuronal signals and then transmit the sound information to the brain. Sound-induced cochlea vibration takes the form of a travelling wave that reaches its maximum at a certain site on the duct and thereafter tapers off. Tonotopy, describes the fact that the site of the maximum vibration along the cochlea is dependent on the frequency of the tone that generates the vibration. High frequencies vibrate the ducts maximally right at the start while low frequency tones generate waves with maxima further along the ducts. Thus, hair cells located at different sites on the ducts are preferentially excited by tones of different frequencies. In the figure the coiled cochlea (blue) has been stretched out to show the principle of tonotopical coding.
This knowledge, unearthed by many decades of intensive basic research, has been successfully exploited to establish the Cochlea Implant. Sound is picked up by a microphone (1), decomposed in its frequencies (2), then relayed through the skin (3) to a bundle of electrodes (4) that stimulate different portions of the cochlea to create the sensation of pitch after relaying the artificially evoked activity towards the brain via the healthy nerve (5). Nowadays ten thousands of patients, amongst them deaf babies that can learn to speak using the device, have been successfully implanted and benefit from it.
At the CIN a device to restore blindness due to diseases of the eye, the so-called Retina Implant is being developed. The basic organizational principle exploited is, as with the cochlea, the orderly representation of sensory signals. In the visual system, this is long known and is called retinotopy. It describes the fact that the visual world is projected by the eye’s optic system in an inverse way onto the retina. Specific problems of the retina implant are the movement of the eye and the delicacy of the retina which is a sheet of dedicated nerve cells only a few hundred microns thick. To safely stimulate the retina photodiodes and stimulation electrodes have been machined into a chip that is flat and can be implanted just below the retina (sub-retinal implant).
The photodiodes capture light and convert it to a current pulse that is applied by the electrodes to neighboring retinal cells. In addition the chip has electronics ‘on board’ to do different processing of the light signals detected. Retinal cells next to the active electrode will be activated and relay the imprinted signal up the visual pathways and toward the brain.
Cochlea and Retina Implant can only alleviate sensory disturbances that are caused by disease of the sensor organ (cochlea or retina). However, neurology is confronted with many sensory disturbances that are due to brain lesion rather than problems with the sense organ. In these cases, only neuroprostheses that imprint sensory signals directly to the brain can be helpful. Fortunately, many sensory areas of the neocortex are topically organized. For auditory, visual and tactile signals this means that topically ordered signals in the sense organ are projected to the neocortex via ascending neuronal pathways in a topical fashion. Therefore, the primary sensory cortices present the sensory surface in a highly ordered fashion and may be amenable for future approaches to imprint detailed sensory information.
The problem of central neuroprostheses is that the neurons to be activated are highly interconnected, thus their activation depends on what else is going in such a network at the time of stimulation. Researchers at the CIN therefore explore effects of cortical microstimulation in different functional contexts; Methods to record cortical activity and adapt microstimulation according to the state of the cortex are being explored as one strategy to cope with the problems of imprinting information into highly feedback coupled networks.