neuralnetworks and expert systems provide different ways to reduce the programming effort required to build complex systems. Adaptive neuralnetworks are programmed merely by training them with examples. Rule-based e...
详细信息
A real-time adaptive scheme is introduced to detect and track moving objects under noisy, dynamic conditions including moving sensors. This approach integrates the adaptiveness and incremental learning characteristics...
详细信息
作者：
Gove, Robert J.Texas Instruments
Incorporated Computer Science Center Speech and Image Understanding Laboratory P.O. Box 655474 MS 238 DallasTX75265 United States
This paper presents a detailed description and a comparative analysis of the algorithms used to determine the position and orientation of an object in real-time. The exemplary object, a freely moving gold-fish in an a...
详细信息
This paper presents a detailed description and a comparative analysis of the algorithms used to determine the position and orientation of an object in real-time. The exemplary object, a freely moving gold-fish in an aquarium, provides "real-world" motion, with definable characteristics of motion (the fish never swims upside-down) and the complexities of a non-rigid body. For simplicity of implementation, and since a restricted and stationary viewing domain exists (fish-tank), we reduced the problem of obtaining 3D correspondence information to trivial alignment calculations by using two cameras orthogonally viewing the object. We applied symbolic processing techniques to recognize the 3-D orientation of a moving object of known identity in real-time. Assuming motion, each new frame (sensed by the two cameras) provides images of the object's profile which has most likely undergone translation, rotation, scaling and/or bending of the non-rigid object since the previous frame. We developed an expert system which uses heuristics of the object's motion behavior in the form of rules and information obtained via low-level imageprocessing (like numerical inertial axis calculations) to dynamically estimate the object's orientation. An inference engine provides these estimates at frame rates of up to 10 per second (which is essentially real-time). The advantages of the rule-based approach to orientation recognition will be compared other pattern recognition techniques. Our results of an investigation of statistical pattern recognition, neuralnetworks, and procedural techniques for orientation recognition will be included. We implemented the algorithms in a rapid-prototyping environment, the TI-Explorer, equipped with an Odyssey and custom imaging hardware. A brief overview of the workstation is included to clarify one motivation for our choice of algorithms. These algorithms exploit two facets of the prototype imageprocessing and understanding workstation - both low-level
A parallel language has to match or reflect the hardware underneath to use these resources efficiently. Though every parallel language has to have some kind of parallel machine model, no existing language states this ...
详细信息
Describes a multilayer pipelined digital architecture suitable for the implementation of large neural nets (LN) for vision applications. It can also be used to do some pre-filtering, such as pixel averaging, by settin...
详细信息
Describes a multilayer pipelined digital architecture suitable for the implementation of large neural nets (LN) for vision applications. It can also be used to do some pre-filtering, such as pixel averaging, by setting weight values appropriately. A 1024 node processor with a clock rate of 10 MHz can operate on an input vector consisting of 32*32 8 bit pixels in 102 mu s. It can therefore process a 256*256 pixel frame in about 6.5 ms, which is well within the requirement for real-time processing, given a frame update rate of 20 ms. The architecture is currently being assessed to see how it can be extended to perform other imageprocessingapplications, and other NN training algorithms.< >
Simple three-layer perceptrons with linear units working in auto-association with a reduced number of hidden units are applied to the task of digitized image compression. First, an algorithm developed using several mu...
详细信息
Simple three-layer perceptrons with linear units working in auto-association with a reduced number of hidden units are applied to the task of digitized image compression. First, an algorithm developed using several multilayer perceptrons in competition for the coding of a TV-image is explained. A theoretical interpretation in terms of principal component analysis is also developed. Then, a study of its performances at different bit-rates is presented. This leads to an extension of the algorithm in image segmentation through texture analysis. Next, the results are compared with conventional methods: the optimal stationary process known as Karhunen-Loeve transform, and an algorithm often proposed for real-time applications, the discrete cosine transform. Finally, an estimation of hardware complexity in the case of real-time television is presented.< >
PCA analysis is similar to neuralnetworks in data compression of segments that have been 'seen', but is superior in compressing 'unseen' images. The difference between 'seen' and 'unseen&#...
详细信息
PCA analysis is similar to neuralnetworks in data compression of segments that have been 'seen', but is superior in compressing 'unseen' images. The difference between 'seen' and 'unseen' images with respect to tsse (total sum square error) is more pronounced in 32 by 16 pixel segments than 8 by 8 segments in PCA compression. This suggests that the dimensional reduction is more consistent in the smaller segments. However the tsse is lower for 'seen' segments in the larger segment PCs. artificialneuralnetworks also seem to generalise less readily on larger segments. Since the time taken for neural network compression is about an order of magnitude higher than PCA, and PCA is more repeatable in terms of the error magnitude, and produces lower error for 'unseen' segments, it would seem preferable to use PCA analysis than neural network methods to produce the reduced dimensional input to a diagnostic network.< >
Describes the application of an artificialneural network, the multi-layer perceptron, to the problem of extracting the position of the eyes from a head and shoulders type image. The multi-layer perceptron is describe...
详细信息
Describes the application of an artificialneural network, the multi-layer perceptron, to the problem of extracting the position of the eyes from a head and shoulders type image. The multi-layer perceptron is described and various architectures of network are considered. The nature of the training data required for the network is discussed, including the size of the input image and the use of whole or partial images for training. Results on both still and moving pictures have been obtained. Attempts have been made to extend the process to a generalised solution which will detect the position of the eyes in any head and shoulders image. The most successful network can be trained on five frames of a moving sequence, and is then able to track the eyes through the rest of the sequence with hardly any problems.< >
作者：
Schoonees, J.A.CSIR
Div for Microelectron & Commun Technol Pretoria S Afr
An introduction to artificialneural network models is presented, along with an overview of their practical application and potential applications in signal processing. Successful neural network implementations are de...
详细信息
ISBN:
(纸本)0879427094
An introduction to artificialneural network models is presented, along with an overview of their practical application and potential applications in signal processing. Successful neural network implementations are described and their performances are compared to those of more traditional signal processing implementations. The Hopfield net, self-organizing feature maps, and the multilayer perceptron are reviewed. Implementation of neural nets in speech synthesis, speech recognition, target identification, imageprocessing, pattern matching, error-correction coding, and neurocomputing are reported. Several ICs currently in production are briefly mentioned.
Models of objects stored in memory have been shown to be useful for guiding the processing of computer vision systems[1,3,7,8,19,23]. A major consideration in such systems, however, is how stored models are initially ...
详细信息
Models of objects stored in memory have been shown to be useful for guiding the processing of computer vision systems[1,3,7,8,19,23]. A major consideration in such systems, however, is how stored models are initially accessed and indexed by the system[17,19,26]. As the number of stored models increases, the time required to search memory for the correct model becomes high. Parallel distributed, connectionist, neuralnetworks1 have been shown to have appealing content addressable memory properties[2,4,5,9,15]. This paper discusses an architecture for efficient storage and reference of model memories stored as stable patterns of activity in a parallel, distributed, connectionist, neural network. The emergent properties of content addressability and resistance to noise are exploited to perform indexing of the appropriate object centered model from image centered primitives. The system consists of three network modules each of which represent information relative to a different frame of reference. The model memory network is a large state space vector where fields in the vector correspond to ordered component objects and relative, object based spatial relationships between the component objects. The component assertion network represents evidence about the existence of object primitives in the input image. It establishes local frames of reference for object primitives relative to the image based frame of reference. The spatial relationship constraint network is an intermediate representation which enables the association between the object based and the image based frames of reference. This intermediate level represents information about possible object orderings and establishes relative spatial relationships from the image based information in the component assertion network below. It is also constrained by the lawful object orderings in the model memory network above. The system design is consistent with current psychological theories of recognition by component[6]. I
暂无评论