Adaptive Thresholding for Object Detection in Low Bitrate Video
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low bitrate video compression artifacts distort image features, leading to lower detection scores and increased false negatives or false positives in computer vision applications like object detection and recognition, due to the use of models trained on high-quality images.
Innovation Solution
Implementing adaptive thresholding for object detection in low bitrate video streams by using high quality I-frames for reliable object detection and translating spatial indicators from I-frames to P-frames, with different detection thresholds applied within and outside regions of interest in P-frames to improve detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If video is compressed to low bitrate to reduce bandwidth, then bandwidth consumption is reduced, but image quality deteriorates causing detection accuracy to decrease
Solution Approach 1:
The patent segments the video processing into two distinct paths: (1) full-resolution frames are processed for object detection to maintain accuracy, and (2) downsampled frames are used for motion estimation and compression. This segmentation allows the system to achieve low bitrate transmission while preserving detection accuracy by preventing compression artifacts from corrupting the detection-critical full-resolution data.
Solution Approach 2:
The patent introduces downsampled frames as an intermediary element that serves dual purposes: they enable motion estimation and compression operations at low computational cost, while the full-resolution frames serve as the ultimate reference for accurate object detection. The downsampled frames act as a mediator that facilitates compression without directly impacting detection quality.
2Measurement precision
If compression artifacts are reduced by using higher bitrate, then detection accuracy improves, but bandwidth consumption increases
Solution Approach 1:
The patent separates compression operations from detection operations by applying compression only to downsampled frames used for motion estimation, while full-resolution frames remain uncompressed and dedicated to object detection. This segmentation eliminates the need to transmit high-bitrate video, reducing bandwidth consumption while preserving detection accuracy.
Solution Approach 2:
The patent applies different quality levels to different processing streams: high quality (full resolution) is maintained for detection-critical operations, while lower quality (downsampled) is sufficient for motion estimation and compression. This local quality differentiation optimizes bandwidth usage without compromising detection performance.
3Productivity
If object detection is performed on low quality decompressed video, then processing speed is maintained, but false negatives and false positives increase
Solution Approach 1:
The patent segments the detection process by performing motion estimation on downsampled frames (maintaining speed) while performing object detection on full-resolution frames (ensuring reliability). This segmentation allows the system to achieve both high processing speed and high detection reliability simultaneously.
Solution Approach 2:
The patent performs preliminary motion estimation and region-of-interest identification on downsampled frames before conducting the actual object detection on full-resolution frames. This preliminary action on lower-quality data speeds up the process by narrowing down search areas, while the final detection on high-quality data ensures reliability.
Data Source
AI summary
Techniques related to applying computer vision to decompressed video are discussed. Such techniques may include generating a region of interest in an individual video frame by translating spatial indicators of a first detected computer vision result from a reference video frame to the individual video frame and applying a greater threshold within the region of interest than outside of the region of interest for computer vision evaluation in the individual frame.