Overhead Video Target Detection Using Line Segments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video surveillance systems face challenges in accurately detecting and tracking individual targets in overhead camera views, often misidentifying groups of people as a single 'blob' and failing to distinguish between humans and inanimate objects, especially in residential and commercial settings.
Innovation Solution
A video surveillance system that uses a combination of moving pixel detection, line segment extraction, and tracking filters to differentiate between multiple objects in tight groups and distinguish humans from inanimate objects, employing techniques like three-frame differencing and Kalman filters to accurately count and verify individuals in various settings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional blob-based detection is used, then processing is simplified, but the ability to count individual targets is lost
Solution Approach 1:
The patent segments each detected blob into multiple potential target instances by analyzing spatial relationships and motion patterns. Instead of treating the entire blob as a single target, the system divides it into candidate targets based on detected line segments and motion vectors, allowing individual counting while maintaining simplified blob-based processing architecture.
Solution Approach 2:
The patent introduces a temporal dimension by using motion information from multiple frames to distinguish between single targets and groups. By analyzing motion coherence across time, the system can identify whether members of a blob are moving together (indicating a group) or independently (indicating separate targets), enabling accurate counting without complex spatial analysis.
2Area of stationary object
If overhead camera views are used, then monitoring coverage is improved, but the ability to distinguish humans from objects is reduced
Solution Approach 1:
The patent utilizes color information as a distinguishing feature between human targets and inanimate objects. By analyzing the color properties of detected objects and comparing them against learned human color characteristics, the system can identify humans even in overhead views where geometric features are less distinctive. This adds a new discrimination dimension without requiring changes to camera geometry.
Solution Approach 2:
The patent introduces motion coherence as an intermediary feature that mediates between the limited visual information available in overhead views and the need for accurate human detection. By analyzing whether detected objects exhibit coherent motion patterns consistent with human movement, the system can reliably identify humans without relying on distinctive geometric or color features alone.
3Productivity
If groups of targets are detected as a single blob, then processing is simplified, but the number of individuals cannot be ascertained
Solution Approach 1:
The patent performs preliminary analysis of motion patterns and spatial relationships within blobs during the detection phase, preparing information that will be used later for individual target identification. By pre-computing motion coherence metrics and spatial distribution characteristics, the system can quickly determine the number of individuals in a group without requiring complex reanalysis, thus maintaining high processing speed while recovering individual information.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A technique for video processing includes: receiving video from an overhead view of a scene; detecting moving pixels in the video; detecting line segments in the video based on detected moving pixels; identifying targets in the video based on the detected line segments; tracking targets in the video based on the identified targets; and managing tracked targets in the video.