Edge-Based Object Recognition for Fast Complex-Background Detection

The patent divides the image processing task into multiple stages: first performing edge detection to identify potential object regions, then applying deep learning only to these extracted regions of interest. This segmentation approach maintains high recognition accuracy while significantly reducing the computational burden compared to processing the entire image through deep learning networks.

The patent performs preliminary edge detection and region extraction before applying deep learning algorithms. By pre-processing the image to identify and isolate potential target objects, the system prepares the data in advance, allowing the deep learning model to focus only on relevant regions and thereby reducing overall computing time while maintaining accuracy.

The patent segments the computational workload between simple edge detection algorithms that can run on resource-constrained edge devices and more complex deep learning models that process only the extracted regions of interest. This division allows edge computing platforms to achieve accurate object recognition without requiring the full computational resources needed for processing entire images through deep learning networks.

The patent extracts regions of interest from the full image using edge detection before feeding them to the deep learning model. This extraction process removes unnecessary background and irrelevant areas, reducing the amount of data that needs to be processed by the computationally intensive deep learning network, thereby lowering device complexity requirements for edge deployment.

The patent segments the image into regions of interest based on edge detection results, then applies deep learning only to these segmented regions. This maintains comprehensive object detection capability by ensuring all potential targets are captured through edge-based region extraction, while improving computational efficiency by limiting deep learning processing to only the necessary portions of the image.

The patent applies deep learning partially to only the extracted regions of interest rather than to the entire image. This partial action approach maintains sufficient object detection comprehensiveness by focusing computational resources on areas most likely to contain target objects, thereby significantly improving computational efficiency without sacrificing detection coverage.