Multi-Camera Object Tracking via Placement Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing monitoring systems struggle to accurately track objects across multiple cameras when feature extraction fails due to varying angles and lighting conditions, leading to inconsistent feature recognition and potential loss of object tracking.
Innovation Solution
A monitoring device that receives videos from multiple cameras, extracts feature information, and uses camera placement data to identify and correct capturing time information, ensuring accurate object tracking by estimating the object's presence in videos where features are not extracted, and specifying the correct camera for object detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature extraction is performed on videos from multiple cameras, then object identification accuracy is improved, but tracking reliability deteriorates when feature extraction fails due to varying angles and lighting conditions
Solution Approach 1:
The system performs preliminary actions by storing camera placement information and predicting object positions in advance. When feature extraction fails, the system uses the stored camera placement data to predict where the object should appear in other camera views, maintaining tracking continuity without requiring successful feature extraction from every camera.
Solution Approach 2:
The system introduces camera placement information as an intermediary element that mediates between failed feature extraction and object tracking. Instead of directly relying on feature matching, the system uses camera positions and angles to predict object locations, serving as a bridge when direct feature-based identification fails.
2Area of stationary object
If multiple cameras are used to track objects, then tracking coverage is improved, but system complexity increases due to the need to manage videos and feature information from multiple sources
Solution Approach 1:
The system extracts and stores only the essential camera placement information (positions and angles) separately from the video data. This extraction of critical geometric data allows the system to manage multiple camera sources without proportionally increasing complexity, as the placement information serves as a reusable reference for all tracking operations.
Solution Approach 2:
The system segments the tracking problem into independent camera views, each processed separately. By treating each camera's video stream and feature extraction as an independent unit and coordinating them through stored placement information, the system manages complexity through modular processing rather than attempting to process all camera data as a single integrated stream.
3Speed
If feature information is used for object identification, then identification speed is improved, but measurement precision deteriorates when lighting conditions vary across different cameras
Solution Approach 1:
The system uses camera placement information as an intermediary to supplement feature-based identification. When lighting variations cause feature recognition inaccuracies, the system refers to the stored camera position and angle data to predict object locations, maintaining precision without sacrificing the speed benefits of feature-based methods.
Data Source
AI summary
A monitoring device (2) identifies an object from videos made by a plurality of cameras (1) including a first camera and a second camera and having a predetermined positional relationship. The monitoring device has a receiving unit (21) configured to receive the videos from the plurality of cameras, a storage unit (22b, 22c) configured to store feature information indicating a feature of the object and camera placement information indicating placement positions of the cameras, and a controller (23) configured to identify the object from the videos based on the feature information. If an object has been identifiable from the video made by the first camera but has been unidentifiable from the video made by the second camera, the controller (23) specifies, based on the camera placement information, the object in the video made by the second camera.