Abstract-The harvest yield in vineyards can vary significantly from year to year and also spatially within plots due to variations in climate, soil conditions and pests. Fine grained knowledge of crop yields would allow viticulturists to better manage their vineyards. The current industry practice for yield prediction is destructive, expensive and spatially sparse -small samples are taken from the vineyards during the growing season and extrapolated to determine overall yield. We present an automated method that uses computer vision to identify and count grape berries. These counts are used to generate per vine estimates of crop yield. Both shape and visual texture are used to detect berries. We demonstrate detection of green berries against a green leaf background. We present crop yield estimation results, with the actual harvest yield as groundtruth for 200 vines (over 450 meters) of two different grape varieties. We calibrate our berry count to yield and find that we can predict yield to within 9.8% of actual crop weight.

We derive a new class of photometric invariants that can be used for a variety of vision tasks including lighting invari Invariants in VisionAppearances of scenes depend on a variety of factors such as lighting geometry and spectrum, scene structure and material properties, medium in which light travels, viewing geometry and sensor properties. Most often, these parameters combine non-linearly to yield an image. Recovering these factors from images is an important problem in vision. Direct estimation of these parameters from a set of images of a scene, however, is generally hard. Photometric invariants provide an intermediate solution to this problem.Invariants usually transform images into a simpler feature space where more accurate algorithms can be developed for the task at hand. To be effective, invariants must satisfy two properties -(a) they must be invariant to certain appearance parameters (say, lighting) and (b) they must have good discriminability with respect to other parameters (say, material properties). Good invariants can be effective for common vision tasks such as illumination invariant recognition, material segmentation and lighting insensitive tracking.There has been a lot of previous work in developing invariants and we will reference a few here. In several instances, the invariants are effective only in special situations. [23,10] has been used to separate diffuse and specular reflections (the diffuse component is invariant to specularities) [12,19,20,21, 4]. However, most of the dichromatic model based works assume either (a) objects with homogeneous reflectance, or (b) specific models for diffuse and specular reflections, or (c) prior knowledge about the diffuse or specular colors, or (d) require at least six light sources [13]. In this work, we do not separate reflection components but are interested in separating material from lighting and shape. Compact representations of objects using images under a large number of lighting conditions have been proposed for lighting invariant recognition [1] and shadow removal [14,5]. However, they often do not have a physical meaning (in terms of object material properties) and therefore hard to use for a problem like material segmentation.In this paper, we present a class of photometric invariants that have the same computational framework for a large set of BRDFs, sensor types as well as the number of image measurements. We begin by deriving a simple image formation model that is valid for the class of "separable" BRDFs. We make no assumption on the exact models of BRDF (Lambertian, Torrence-Sparrow, etc.) in our image formation model. Then, we show how to decompose the observed image measurements into material properties of scene points and the lighting and scene geometry. This is the key idea for creating lighting and shape invariants as well as material invariants. Most of the invariants in this class need changes in illumination or object position between image acquisitions. The stabilities of the invariants increase when the changes in...

No abstract

Free-text database searching is a natural candidate for acceleration by run-time reconfigurable custom computing machines. We describe a fully pipelined search machine architecture for scoring the relevance of textual documents against approximately 100 relevant target words, with provision for limited regular expression matching and error tolerance. An implementation on the SPACE custom computing platform indicates that throughput in the order of 20 megabytes per second is achievable on Algotronix FPGAs i f a locally synchronous design style is adopted and global communications minimized. Partial reconfiguration of the datapath at run-time, in around 3 seconds, serves to maximize the density of data storage on the machine and correspondingly avoid costly input from the environment.

We present Castor, a secure code-update protocol for sensor networks that exploits symmetric cryptoystems. Through a synergistic combination of a one-way hash-chain, two oneway key-chains with the delayed disclosure of symmetric keys, and multiple message authentication codes (MACs), Castor enables untrusted sensor nodes to verify an update's authenticity and guarantees that no correct node will ever install or forward a compromised part of a code-update image. We describe an implementation of Castor that hardens the TinyOS-based update protocol, Deluge, against node compromise. We experimentally compare Castor's computational and communication costs with those of Deluge and with those of a contemporary secure update protocol, Sluice, that uses asymmetric cryptosystems (digital signatures) instead. Our results demonstrate that Castor incurs reasonable overheads as compared to Deluge, and lower resource usage as well as lower end-to-end update latency as compared to Sluice.