Real time Stereo Vision Based On Biologically Motivated Algorithms Using GPU (original) (raw)
Related papers
Analysis of Real-Time Stereo Vision Algorithms On GPU
2011
Dozens of stereo correspondence algorithms whose matching performance has been measured are available, but the trade-off between speed and matching performance of viable realtime stereo has received much less attention. Here, we evaluate five correspondence algorithms(Symmetric Dynamic Programming Stereo ,SemiGlobal Matching,simple block matching, Belief Propagation, and its constant space variant) on a GPU using CUDA and report running time and matching performance. Analysis of these results leads to several insights into the advantages and limitations of the five algorithms -all on a GPU -and SemiGlobal block matching on a conventional CPU.
Comparison of FPGA and GPU implementations of real-time stereo vision
2010
Real-time stereo vision systems have many applicationsfrom autonomous navigation for vehicles through surveillance to materials handling. Accurate scene interpretation depends on an ability to process high resolution images in real-time, but, although the calculations for stereo matching are basically simple, a practical system needs to evaluate at least 10 9 disparities every second -beyond the capability of a single processor. Stereo correspondence algorithms have high degrees of inherent parallelism and are thus good candidates for parallel implementations. In this paper, we compare the performance obtainable with an FPGA and a GPU to understand the trade-off between the flexibility but relatively low speed of an FPGA and the high speed and fixed architecture of the GPU. Our comparison highlights the relative strengths and limitations of the two systems. Our experiments show that, for a range of image sizes, the GPU manages 2 × 10 9 disparities per second, compared with 2.6 × 10 9 disparities per second for an FPGA.
Programmable logic devices, such as Field Programmable Gate Arrays, are well-suited for implementing biologicallyinspired visual processing algorithms and among those algorithms is HMAX model. This model mimics the feedforward path of object recognition in the visual cortex. Methods: HMAX includes several layers and its most computation intensive stage could be the S1 layer which applies 64 2D Gabor filters with various scales and orientations on the input image. A Gabor filter is the product of a Gaussian window and a sinusoid function. Using the separability property in the Gabor filter in the 0° and 90° directions and assuming the isotropic filter in the 45° and 135° directions, a 2D Gabor filter converts to two more efficient 1D filters. Results: The current paper presents a novel hardware architecture for the S1 layer of the HMAX model, in which a 1D Gabor filter is utilized twice to create a 2D filter. Using the even or odd symmetry properties in the Gabor filter coefficients reduce the required number of multipliers by about 50%. The normalization value in every input image location is also calculated simultaneously. The implementation of this architecture on the Xilinx Virtex-6 family shows a 2.83ms delay for a 128×128 pixel input image that is a 1.86X-speedup relative to the last best implementation. Conclusion: In this study, a hardware architecture is proposed to realize the S1 layer of the HMAX model. Using the property of separability and symmetry in filter coefficients saves significant resources, especially in DSP48 blocks.
Fast GPU Accelerated Stereo Correspondence
2015
Many surveillance applications could benefit from the use of stereo cameras for depth perception. While state-of-the-art methods provide high quality scene depth information, many of the methods are very time consuming and not suitable for real-time usage in limited embedded systems. This study was conducted to examine stereo correlation methods to find a suitable algorithm for real-time or near real-time depth perception through disparity maps in a stereo video surveillance camera with an embedded GPU. Moreover, novel refinements and alternations was investigated to further improve performance and quality. Quality tests were conducted in Octave while GPU suitability and performance tests were done in C++ with the OpenGL ES 2.0 library. The result is a local stereo correlation method using Normalized Cross Correlation together with sparse support windows and a suggested improvement for pixel-wise matching confidence. Applying sparse support windows increased frame rate by 35% with m...
Implementation of a Memory-Efficient Stereo Vision Algorithm Based on 1-D Guided Filtering
2014
This paper presents an FPGA implementation of a memory-efficient stereo vision algorithm. Recently, a hardwareoriented stereo vision algorithm using 1-D guided filtering was proposed [1]. Our architecture is based on this algorithm and calculates the depth map in 1-D space, and therefore, the required amount of memory is significantly reduced. To realize high speed processing, we apply a full-pipeline and highly parallel structure. Implemented on FPGA, our design uses an 89 kb memory and achieves a 188 frame per second rate for 384×288 stereo images. The accuracy of the disparity map is not satisfactory; however, it can be improved by a small amount of software processing (8.6 ms) [1]. This result shows that the proposed architecture is highly efficient in terms of the required memory, and its processing speed and accuracy is the same as that of other methods.
Implementation of Stereo Matching Algorithms on Graphics Processing Units
2009
Stereovision is a passive technique for reconstruction of 3D scenes. Unfortunately, depth calculation in this imaging technique is computationally demanding. In this paper we show that stereovision matching algorithms can be efficiently mapped onto modern GPU graphics cards. A number of modifications to depth computations have been proposed that make running them on GPU platforms particularly efficient. Example computed depth maps are shown and time performances of the proposed algorithms are outlined.
Local Stereo Parametric Methodsof Disparity Computation using GPU
2016
Vision is the most important sense for humans and because of this human vision system only we are able to see the 3D world around us with great clarity and are able to find out depth of each and every object. Many Active and Passive depth estimation techniques have been proposed which are capable of estimating depth of real world scene among which one of the passive method, stereo vision has been proven to provide remarkable results. We have used stereo vision technique to estimate depth for a given real world scene. We have done calibration and used non rectified as well as rectified stereo images and then algorithms such as SAD, SSD, NCC, SAD by Derivatives are used for estimating reliable and accurate correspondence match for stereoscopic image pairs. We have used window and aggregation approach to improve the accuracy of disparity map and triangulation method is used to compute depth from disparity space image matrix .The algorithms are implemented in multicore processors using ...