Local and Global Map Tracking With Kalman Duplicate Resolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection and classification systems face challenges in achieving robust, accurate, and adaptable object detection, classification, and localization, particularly in complex scenarios, due to limitations in memory, processing, and resource constraints, leading to issues like false detections, duplicate detections, and localization errors.
Innovation Solution
A multi-modal, multi-camera system integrated with a first logic device that fuses sensor inputs using a trained neural network, Kalman filtering, and a local and global object tracking module to detect, classify, and localize objects, with a human-in-the-loop interface for validation and refinement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a trained neural network is loaded on a server system to classify images, then object detection accuracy is improved, but system complexity and resource requirements increase
Solution Approach 1:
The system divides object detection into two independent parts: a server system that provides trained neural network models and a client system that performs actual image classification. This segmentation allows the complex neural network to be accessed remotely without requiring the client to host the entire model, thus improving detection accuracy while managing system complexity through distributed architecture.
Solution Approach 2:
A server system acts as an intermediary between the client imaging devices and the trained neural network. The server receives images from clients, processes them through the neural network, and returns classifications. This intermediary approach enables accurate object detection without requiring complex computational resources on the client side, resolving the contradiction between accuracy and system complexity.
2Speed
If a trained neural network is loaded on an imaging system itself for real time operation, then real time object detection is achieved, but memory and processing resource constraints are exceeded
Solution Approach 1:
The trained neural network is extracted from the client imaging system and placed on a separate server. The client system only performs simple image acquisition and transmission, while the server handles the computationally intensive neural network processing. This extraction allows real-time operation to be maintained through network communication while keeping the resource-constrained client system lightweight.
3Device complexity
If traditional rules-based image processing software is used, then system simplicity is maintained, but detection accuracy for complex objects is insufficient
Solution Approach 1:
The system combines the simplicity of rules-based processing on the client side with the power of neural networks on the server side. The client performs simple tasks like image capture and basic preprocessing, while the server handles complex classification. This universal approach allows the same architecture to work for both simple and complex detection tasks by leveraging the appropriate processing capability at the right location.
4Productivity
If object detection is performed without Kalman filtering, then processing speed is maintained, but false detections and duplicate detections occur
Solution Approach 1:
Kalman filtering implements a feedback mechanism that uses historical object position information to predict current positions and correct measurement errors. The filter continuously updates object trajectories based on previous detections and current sensor data, effectively eliminating false and duplicate detections while maintaining processing speed through efficient mathematical operations.
Data Source
AI summary
A detection device, such as an unmanned vehicle, is adapted to traverse a search area and generate sensor data associated with objects that may be present in the search area. The generated sensor data is used by a system including object detection inference models configured to receive the sensor data and output object data, a local object tracker configured to track detected objects in a local map, and a global object tracker configured to track detected objects on a global map. The local object tracker is configured to fuse object detections from the object detection inference models to identify locally tracked objects, and a Kalman filter processes frames of fused object data to resolve duplicates and/or invalid object detections. The global object tracker includes a pose manager, configured to track global objects in the global map and update the pose based on a map optimization process. User-in-the-loop processing includes a user interface for displaying and manual editing of detected object data.