Two-Stage Object Detection for Embedded Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection methods for embedded devices are inefficient due to the large search space required for detecting objects of different scales, which is daunting for resource-limited systems.
Innovation Solution
A two-stage object detection scheme is implemented, using a simple detector to prune easy false positives and a more powerful detector to improve true positive proposals, with the second stage applied only to a reduced set of search locations, and incorporating a diamond-shaped region for non-maximum suppression.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a complex detector is applied over the entire region of interest to detect objects of different scales, then detection accuracy is improved, but computational complexity and processing time increase significantly
Solution Approach 1:
The detection process is divided into two stages: a first detection stage that processes the entire image to identify candidate regions, and a second detection stage that processes only those candidate regions with higher precision. This segmentation reduces the overall computational complexity while maintaining detection accuracy for objects of different scales.
Solution Approach 2:
The first detection stage performs preliminary processing to identify and filter candidate regions before the second detection stage applies more complex analysis. This preliminary action reduces the search space for the complex detector, decreasing computational complexity while preserving detection accuracy.
2Measurement precision
If a complex detector is applied over the entire region of interest to detect objects of different scales, then detection accuracy is improved, but processing time increases significantly
Solution Approach 1:
The detection process is divided into two stages: a first detection stage that quickly identifies candidate regions across the entire image, and a second detection stage that processes only those candidates with higher computational complexity. This segmentation significantly reduces total processing time while maintaining detection accuracy.
Solution Approach 2:
The first detection stage performs preliminary filtering to identify promising candidate regions before the second detection stage applies more time-consuming complex analysis. This preliminary action eliminates the need to process entire images with complex detectors, reducing processing time while preserving accuracy.
3Productivity
If a simple detector is used to reduce computational complexity, then processing speed is improved, but detection accuracy deteriorates
Solution Approach 1:
The detection system uses a simple detector for the first stage to quickly identify candidate regions, then applies a complex detector in the second stage to those candidates. This two-stage segmentation maintains high processing speed in the first stage while achieving high detection accuracy in the second stage.
Solution Approach 2:
The simple detector in the first stage performs preliminary identification of candidate regions, which are then refined by the complex detector in the second stage. This preliminary action allows the system to maintain high processing speed while achieving accurate detection results.
Data Source
AI summary
An apparatus comprises a detector and a processor. The processor may be configured to perform a two-stage object detection process utilizing the detector circuit. The detector circuit may be configured to implement a simple detection stage and a complex detection stage. In the simple detection stage, a first type of object detection is applied to each image in a search space and a number of candidates are identified by applying a first non-maximum suppression technique. In the complex detection stage, a second type of object detection is applied to the candidates identified by the simple detection stage.