Dual-Stage Object Detection for Embedded Vision Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection techniques face challenges in achieving fast and accurate detection on devices with limited calculating resources, particularly when dealing with large numbers of detection targets in images, as deep learning algorithms require significant computational power and are not suitable for embedded devices.
Innovation Solution
A dual-detection system is implemented, where a first detection unit identifies candidate regions using a low-load algorithm, and a second detection unit performs accurate detection only on integrated regions selected from these candidates, utilizing a more computationally intensive but accurate algorithm like deep learning, thereby reducing the overall calculation load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning algorithm is used for object detection, then detection accuracy is improved, but calculation load increases making it unsuitable for embedded devices
Solution Approach 1:
The patent segments the detection process into two stages: first using a low-load algorithm to detect candidate regions, then applying the high-accuracy deep learning algorithm only to those candidate regions. This segmentation allows the system to achieve high detection accuracy while reducing overall calculation load, making it suitable for embedded devices with limited computing resources.
2Use of energy by moving object
If low-load algorithm is used for detection, then calculation load is reduced, but detection accuracy decreases
Solution Approach 1:
The detection process is divided into two sequential stages: a first detection unit uses a low-load algorithm to identify candidate regions, and a second detection unit applies a high-accuracy algorithm (such as deep learning) specifically to those candidate regions. This segmentation allows the system to achieve high detection accuracy while maintaining low overall calculation load.
3Measurement precision
If all candidate regions are processed by highly accurate algorithm, then detection accuracy is maintained, but processing speed decreases due to large number of regions
Solution Approach 1:
The patent extracts and processes only the most promising candidate regions identified by the first detection unit. By applying the highly accurate second detection algorithm only to these extracted candidate regions rather than all possible regions, the system maintains high detection accuracy while significantly improving processing speed.
4Reliability
If region integration is performed on all detected regions, then false detections are reduced, but number of integrated regions increases leading to slower processing
Solution Approach 1:
The patent applies region integration selectively only to candidate regions identified by the first detection unit, rather than performing integration on all possible regions. This local application of integration maintains detection reliability by reducing false positives in candidate regions while preserving processing speed by avoiding unnecessary integration operations elsewhere.
Data Source
AI summary
An object detection device for detecting a target object from an image, includes: a first detection unit configured to detect a plurality of candidate regions in which the target object exists from the image; a region integration unit configured to determine one or more integrated regions based on the plurality of candidate regions detected by the first detection unit; a selection unit configured to select at least a part of the integrated regions; and a second detection unit configured to detect the target object from the selected integrated region using a detection algorithm different from a detection algorithm used by the first detection unit.