Object Detection Device Using Previous Frame Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection techniques using two-stage detection processes are inefficient due to repeated erroneous detections by preceding detectors, leading to increased processing time, especially with stationary cameras where non-target objects are detected similarly.
Innovation Solution
An object detection device employing a first detector to identify candidate areas, a determiner to select target areas based on previous frame detection information, and a second detector with a different algorithm for accurate detection, reducing redundant processing by excluding similar candidate areas from the second detection stage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a two-stage detection process is used with a preceding detector and a subsequent detector, then detection accuracy is improved, but processing time increases
Solution Approach 1:
The system performs preliminary detection using a first detector to identify candidate areas, then uses detection information from previous frames to pre-determine which candidate areas should be excluded from second-stage detection. This preliminary filtering action reduces the workload of the second detector while maintaining accuracy.
Solution Approach 2:
The system uses detection information from previous frames as feedback to adjust the current detection process. By storing and referencing past detection results, the system can avoid re-detecting areas that have been correctly identified or erroneously detected in previous frames, thereby reducing redundant processing.
2Device complexity
If a stationary camera is used for object detection, then detection coverage is simplified, but erroneous detection of non-target objects increases
Solution Approach 1:
The system performs preliminary detection with a first detector that can identify target objects even in complex stationary camera scenarios. By establishing reliable candidate areas in advance, the system prepares to filter out erroneous detections before they reach the second detection stage.
Solution Approach 2:
The system uses feedback from previous frame detection information to identify and exclude areas where erroneous detections occurred. This feedback mechanism allows the system to learn from past mistakes and avoid repeating them, thereby improving reliability in stationary camera applications.
3Quantity of substance
If the subsequent detector processes all candidate areas, then detection completeness is maintained, but processing time increases due to redundant detection
Solution Approach 1:
The system extracts and removes candidate areas from the second detection stage based on detection information from previous frames. By taking out areas that have been correctly identified or erroneously detected in previous frames, the system reduces redundant processing while maintaining completeness through selective inclusion of only necessary areas.
Solution Approach 2:
Instead of processing all candidate areas through the second detector, the system applies partial action by selectively processing only those candidate areas that require second-stage verification. This partial processing approach maintains sufficient detection completeness while significantly reducing processing time.
Data Source
AI summary
An object detection device for detecting a predetermined object from an image includes a first detector that detects one or more candidate areas including the predetermined object from the image, a determiner that determines a target area from the one or more candidate areas detected by the first detector, a second detector that detects the predetermined object in the target area with a detection algorithm different from a detection algorithm used by the first detector, and a storage that stores detection information indicating a detection result obtained by the second detector for the target area. The determiner determines the target area from the one or more candidate areas based on the detection information about a frame being at least one frame preceding a current frame.