Motion Detection Using Delta SAD and Temporal Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional video cameras have limited processing power, leading to inaccurate motion detection in video surveillance due to reliance on video parameters like motion vectors and sum of accumulated differences (SAD), which can result in false positives and are susceptible to encoding noise.
Innovation Solution
A video camera method that calculates differences in sum of accumulated differences (ΔSAD) for macroblocks, identifies motion-generative macroblocks, forms bounding boxes around candidate-motion regions, and applies a temporal filter to distinguish true motion from artifacts, using the H.264 encoding standard and thresholds to validate motion across successive frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional video cameras use motion vectors and sum of accumulated differences (SAD) for motion detection, then motion detection capability is provided, but false positives occur due to encoding noise and limited processing power
Solution Approach 1:
The video frame is divided into macroblocks, and motion detection is performed at the macroblock level rather than at the entire frame level. This segmentation allows for more precise identification of motion regions and reduces false positives by analyzing local changes individually. The patent applies this by calculating ΔSAD for each macroblock and identifying motion-generative macroblocks based on threshold criteria.
Solution Approach 2:
The patent introduces a new parameter ΔSAD (difference of sum of accumulated differences) to detect motion. Instead of using traditional SAD values directly, the system calculates the change in SAD between successive frames, which provides a more reliable indicator of actual motion. This parameter transformation helps filter out false positives caused by encoding noise while maintaining detection capability.
2Device complexity
If video cameras have limited processing power, then device complexity is reduced, but motion detection reliability deteriorates due to inaccurate detection
Solution Approach 1:
The patent applies partial action by focusing motion detection on only the necessary macroblocks that exceed the flooding threshold, rather than processing the entire frame uniformly. This selective approach reduces processing complexity while maintaining reliability. The system identifies motion-generative macroblocks through a two-stage threshold process (motion-threshold and flooding-threshold), which efficiently filters false positives without requiring excessive computational resources.
Solution Approach 2:
The system performs preliminary filtering by first identifying macroblocks with ΔSAD greater than the motion-threshold value, then further filtering those with ΔSAD greater than the flooding-threshold value. This preliminary action before final motion determination reduces the number of candidates that need full analysis, thereby reducing overall processing complexity while improving reliability.
3Ease of operation
If traditional motion detection methods are used, then ease of operation is maintained, but false motion detection occurs due to reliance on encoding parameters
Solution Approach 1:
The patent introduces ΔSAD as an intermediary parameter between the raw video data and the final motion detection decision. This intermediary transformation layer filters out the influence of encoding noise while preserving motion information. The system uses ΔSAD as a mediator to make more accurate motion determinations without changing the basic operation flow, thus maintaining ease of operation while improving precision.
Data Source
AI summary
A motion detection method of a video camera includes calculating differences for a sum of accumulated differences (ΔSAD) for corresponding macro blocks in successive frames of video. The method may include identifying motion-generative macroblocks and regions of connected macroblocks. Candidate-motion regions may be defined in a frame and a bounding box may be formed around each of the candidate-motion regions. If corresponding bounding boxes in a plurality of successive frames have substantially the same location, size, and shape in the plurality of successive frames, and if the plurality of successive frames are in a temporal window of successive frames, then the corresponding bounding boxes may indicate motion.