Suspicious Activity Detection via Frame-to-Frame Feature Association
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing security monitoring systems in sensitive locations, such as airports and military bases, often generate false alarms due to occlusions and object merging, leading to wasted resources and potential oversight of real threats.
Innovation Solution
A method that detects low-level feature sets in a sequence of image frames and determines frame-to-frame associations to identify suspicious activities, reducing false alarms by modeling these associations as a directed graph and applying user-definable criteria to distinguish between genuine and trivial movements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If single-track object tracking is used for monitoring, then the system is simple to operate, but false alarms increase due to occlusions and object merging
Solution Approach 1:
The patent segments the tracking problem by dividing objects into multiple feature sets (e.g., head, body, limbs) that can be independently detected and tracked. This segmentation allows the system to maintain reliability during occlusions by tracking individual feature sets rather than requiring complete object visibility, thereby resolving the contradiction between operational simplicity and detection reliability.
Solution Approach 2:
The patent transitions from single-track 2D tracking to multi-track association across multiple dimensions by establishing frame-to-frame associations of feature sets. This dimensional expansion allows the system to track objects through occlusions by maintaining multiple potential tracks and associating them across frames, improving reliability while maintaining operational simplicity through automated association algorithms.
2Measurement precision
If high sensitivity is used to detect suspicious activities, then detection accuracy improves, but false alarm rate increases
Solution Approach 1:
The patent implements feedback through frame-to-frame association of feature sets, where detection results from previous frames inform current frame analysis. This feedback mechanism allows the system to maintain high sensitivity by confirming suspicious activities across multiple frames rather than triggering on single-frame detections, thereby reducing false alarms while preserving detection accuracy.
Solution Approach 2:
The patent applies preliminary action by pre-establishing association rules and criteria for suspicious activity detection before analysis. By defining what constitutes suspicious behavior in advance (e.g., loitering patterns, perimeter breaches), the system can maintain high detection sensitivity while reducing false alarms through pre-programmed decision criteria that filter out trivial occurrences.
3Reliability
If reduced sensitivity is used to reduce false alarms, then false detection rate decreases, but real security threats may be overlooked
Solution Approach 1:
The patent segments threat detection into multiple independent feature set tracks, allowing the system to maintain reduced overall sensitivity while preserving individual track sensitivity. By analyzing multiple segmented feature sets rather than requiring complete object detection, the system can lower the threshold for triggering alerts without increasing false alarms from trivial occurrences, thus maintaining threat detection capability while reducing false positives.
Data Source
AI summary
A method and apparatus for detecting suspicious activities is disclosed. In one embodiment at least one low-level feature set is detected in a plurality of frames of a sequence of scene imagery. The frame-to-frame associations of the detected low-level feature set are determined, and suspicious activities is identifying on the bases of these frame-to-frame associations.