The vertebrate eye (figure 1) has provided inspiration for the design of conventional cameras, which consist of a planar image sensor in the focal plane of single-aperture optics (single or multiple co-axial lenses). Their purpose is to provide a faithful rendering of the visual world that can be processed by image analysis algorithms for a large variety of purposes, especially for object recognition. However, such vision systems require complex and fast computation in order to extract motion-related information from a sequence of high-resolution images.
The insect compound eye (figure 2), instead, consists of a curved array of micro-lenses, each conveying photons to a separate set of one or more photoreceptors. Although the compound eye design offers a comparatively lower resolution than the vertebrate eye, it is very efficient for local and global motion analysis over a large field of view (FOV), making it an excellent sensor for accurate and fast navigation in 3D dynamic environments. Furthermore, compound eyes take several shapes and curvatures to fit the head and viewing directions of very different types of insects while offering the same functionality. Vertebrate eye (left) and insect compound eye (centre) with detailed view of a patch of ommatidia (right).
Artificial vision systems inspired upon the insect compound eye will be a radically different alternative to conventional cameras and will provide more efficient visual abilities for embedded applications that require motion analysis in low-power and small packages; artificial compound eyes could also adapt to different shapes and curvatures to fit the application requirements. However, the design of artificial compound eyes presents several technological and scientific challenges because it drastically departs from the design of conventional cameras for what concerns the components, fabrication procedures, packaging, and visual processing.
The grand goal of this project is the design, prototyping, programming, and validation of fully functional artificial compound eyes, which will be composed of micro-lens arrays integrated with adaptive photoreceptors made of analog Very-Large-Scale-Integration (aVLSI, cf. Liu et al., 2003) circuits on flexible electronic substrates. The output of the artificial compound eyes will be processed by vision filters implemented in encapsulated programmable devices, such as microcontrollers or Field Programmable Gate Arrays (FPGA) for fast extraction of motion-related information. We call these integrated vision sensors CURVed Artificial Compound Eyes (CURVACE).
Compared to conventional cameras, the proposed CURVACE will offer much larger field-of-view, nearly infinite depth-of-field (no focusing needed), higher sensitivity, no image blurring and off-axis aberrations because the distance between the optical surface and the photoreceptors will be constant over the entire fieldofview and because each optical channel will work under perpendicular light incidence for its individual viewing direction. In comparison with classical cameras where focal length, spatial resolution and field of view are intimately coupled, a curved compound eye allows the use of different focal length for the same field of view. Furthermore, the curved shape of the artificial compound eyes will offer space within the convexity for embedding processing units, battery, wireless communication, and inertial sensors, such as accelerometers and rate gyroscopes, which will be used for motion-related computation. Instead, in conventional cameras these components must be packaged separately because the space between the convex lens and the planar image sensor must be transparent.
In order to reach the grand goal of artificial compound eyes and the measurable objectives listed above, we will take leverage from the completely novel combination of micro-optical fabrication technologies, adaptive photoreceptors in aVLSI chips, micro-electronics on bendable substrates, and motion detection on bendable imaging surfaces. The coupling of adaptive aVLSI photoreceptors and of compound optics will be particularly promising because it will provide a large dynamic range and will optimally exploit the specific topologies of the two layers. This is because the microlenses will focus all incoming photons only onto the light-sensitive areas of the aVLSI surface circuit while the aVLSI areas occupied by adaptive circuitry will fall between adjacent micro-lenses. In addition to the development of a library of motion-related filters for compound eyes, we will also study and develop novel visual filters that self-adapt to the changing curvatures of the surfaces where the compound eyes will be attached. Furthermore, will explore bio-inspired principles of active vision, for example by applying specific micro-movements that will greatly increase visual acuity despite the coarse resolution of the compound eye, making CURVACE the first micropanoramic eye endowed with hyper acuity.
Curved vision systems and compound eyes have recently raised the interest of the scientific and technology community, but none of the prototypes and fabrication procedures presented so far managed to achieve all our objectives, as we explain in the next section. In order to succeed in our objective, we will explore a novel and unique approach to the fabrication of curved artificial compound eyes that will consist of developing flat patches of artificial ommatidia that will be flexed after fabrication in order to maintain perfect alignment of the optics and photoreceptors that compose each single ommatidium (figure 2), thus enabling the desired optical and imaging properties at the core of the first fully functional artificial compound eyes. This procedure will also allow us to fabricate various types of curved compound eyes. In particular, we will develop and assess four families of CURVACEs: cylindrical, active, spherical, and tape compound eyes. This strategy will also allow us to incrementally tackle the technical and scientific challenges that we will encounter during the project and at the same time develop different prototypes that will suit the needs of different applications.
Another important objective of the CURVACE design process is to make the resulting prototypes reliable and replicable so that they could rapidly spread in the research and educational community, thus triggering a new wave of research into insect-inspired vision, which in the future may have an impact in several ICT applications. For this reason, we will integrate CURVACE prototypes with programmable devices, which should encourage the development and sharing of visual software libraries. Furthermore, we will design the readout electronics so to make them compatible with other types of processing devices, such as neuromorphic chips, which could be directly interfaced to the aVLSI adaptive imagers at the place of the programmable device.