Researchers from Jadavpur University and Intel have developed StEdge, a stochastic computing-based hardware accelerator for real-time edge detection. The architecture demonstrates significant reductions in FPGA slice LUT usage and on-chip power consumption compared to traditional Sobel and Prewitt implementations.
This development addresses the critical power-efficiency bottleneck in real-time video processing for resource-constrained devices. By utilizing stochastic computing to simplify gradient calculations, the architecture allows for high-performance edge detection without the heavy silicon footprint of traditional multipliers. For the streaming ecosystem, this technology facilitates more sophisticated on-device computer vision and metadata extraction in low-power IoT sensors and portable cameras. As edge computing demands grow, the industry should monitor whether this stochastic approach is adopted for more complex encoding tasks or higher-resolution 4K vision pipelines.
The push for 8K edge video processing continues to drive innovation in hardware-accelerated detection and analysis, alongside new physical edge AI hardware designed for sensor-level inference. Future implementations may also leverage multi-stream processing to handle edge AI requirements, potentially integrating AI metadata scanning for automated content analysis.
Researchers from Jadavpur University and Intel have introduced StEdge, a stochastic computing-based hardware accelerator for real-time edge detection. By replacing arithmetic-intensive operations with lightweight logic gates, the architecture achieves a 28% reduction in power consumption. This innovation enables high-performance computer vision on resource-constrained IoT sensors and portable cameras.
StEdge is a stochastic computing-based architecture developed by researchers from Jadavpur University and Intel designed to perform real-time edge detection with significantly reduced computational overhead.
StEdge reduces total on-chip power consumption by 17% and overall edge detection power by 28% compared to traditional Sobel implementations.
The architecture supports both Sobel and Prewitt gradient operators through a unified, configurable datapath and has demonstrated 30 fps performance on a Basys 3 FPGA.
Yes, stochastic Sobel detectors maintained over 80% of the accuracy of deterministic counterparts across all BSD500 benchmark metrics.
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source