VLSI neuroprocessors for video motion detection

Abstract

The system design of a locally connected competitive neural network for video motion detection is presented. The motion information from a sequence of image data can be determined through a two-dimensional multiprocessor array in which each processing element consists of an analog neuroprocessor. Massively parallel neurocomputing is done by compact and efficient neuroprocessors. Local data transfer between the neuroprocessors is performed by using an analog point-to-point interconnection scheme. To maintain strong signal strength over the whole system, global data communication between the host computer and neuroprocessors is carried out in a digital common bus. A mixed-signal very large scale integration (VLSI) neural chip that includes multiple neuroprocessors for fast video motion detection has been developed. Measured results of the programmable synapse, and winner-takes-all circuitry are presented. Based on the measurement data, system-level analysis on a sequence of real-world images was conducted.

A general motion model and spatio-temporal filters for 3-D motion interpretations

“…This coincides with our previous observation in Fig 44. In the extreme case, when st = 1 , we derive the following equation and t-I{x, y, t) = 0 . OX ( 61 ) In equation 61, t is interpreted as the time-of-contact and at that instant, the entire image texture disappears, which is what we observe when the object is too close. It is clear that s measures not only 2-D expansion, but also image gradient evolution and time-tocontact.…”

“…Dedicated VLSI chips Vision Chips: gradient method [73] [105] , correspondence method [28] [104] and biological receptive field design [32] [72] low resolution Non-Vision Chips: analog neural networks [61] digital block matching technique [8] [50] [115] quantized results [69] . computing time-to-contact (and hence obstacle avoidance) [25] and segmentation [99] requirement of tem- The hardware approach uses specialized hardware [34] to achieve real-time performance.…”

“…Another method uses digitized images as input to the VLSI chips to pe orm motion estimation, i.e., there are no on-chip photosensors. One of the two major approaches uses analog neural networks [61] while the other uses digital block matching techniques [8] [50] [115] . This method achieves higher speed and marginally higher resolution than using vision chips.…”

Multi-modal Medical Images Registration Using Differential Geometry and the Hausdorff Distance

“…Differential geometry is helpful in exploiting the curvature and crestlines of the surface of anatomical and physiological structures. Many fast neural network-based implementations are currently available for extracting principal curvatures (Zhou & Chellappa, 1988;Lee et al, 1993), which makes this feature extraction technique more feasible in a real-time clinical environment. Many fast neural network-based implementations are currently available for extracting principal curvatures (Zhou & Chellappa, 1988;Lee et al, 1993), which makes this feature extraction technique more feasible in a real-time clinical environment.…”

Length–speed ratio (LSR) as a characteristic for moving elements real-time classification

“…The classical algorithms of this type are those based on gradient analysis, the most well known being the optical flow model [18][19][20]. The problem with algorithms of this class is that the calculations are again very costly in hardware [21] and, in addition, their implementation in a real-time context requires many simplifications. Other well-known methods of this type are those based on image difference or on accumulated image difference [20]; both methods require a reference image and both are designed for use with a small sequence of images rather than an indefinite sequence.…”