Cognitive Surveillance Robot for Indoor Hazard Detection and Alerts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing surveillance robots lack the ability to effectively detect and respond to indoor hazards and security violations due to limited scene understanding and lack of computational cognitive abilities, particularly in handling indoor objects and anomalies, and are unable to provide continuous 24/7 monitoring.
Innovation Solution
A cognitive surveillance robot system that processes camera inputs using a multi-step pipeline involving image preprocessing, object detection, signage recognition, and semantic attribute extraction, followed by inference and alert triggering using a knowledgebase and reasoning techniques to identify and respond to indoor hazards and security breaches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Duration of action of stationary object
If a mobile robot is deployed for surveillance, then continuous 24/7 monitoring capability is improved, but device complexity and computational requirements increase
Solution Approach 1:
The surveillance system is segmented into modular components: a mobile robot unit with camera for physical patrol, and a separate server system with trained machine learning models for cognitive processing. This segmentation allows the robot to perform continuous monitoring while the complex computational tasks are handled by the server, distributing system complexity across multiple independent modules.
Solution Approach 2:
A communication interface acts as an intermediary between the mobile robot and the server. The robot captures images and transmits them through this intermediary to the server for processing, which then returns alerts or commands. This intermediary layer decouples the robot from the complex processing requirements, enabling continuous monitoring without overloading the mobile device.
2Measurement precision
If computational cognitive ability is added to detect indoor objects and anomalies, then detection accuracy is improved, but processing time and computational resources increase
Solution Approach 1:
Machine learning models are trained in advance on extensive datasets containing various indoor objects, hazards, and anomalies. This preliminary training equips the system with pre-learned knowledge that enables rapid recognition during actual surveillance operations. The heavy computational lifting is done beforehand, allowing fast real-time detection with minimal processing delay.
Solution Approach 2:
The system uses trained machine learning models that are copied from the server to the mobile device for local inference. These pre-trained models contain learned patterns and features that enable accurate object detection without requiring the mobile device to perform complex training computations in real-time, thus maintaining both accuracy and speed.
3Adaptability or versatility
If multiple detection algorithms and models are integrated, then detection capability is improved, but device complexity increases
Solution Approach 1:
The server system is designed as a universal platform that hosts multiple trained machine learning models capable of detecting various types of objects, hazards, and anomalies. This multi-functional server can handle different detection tasks through a unified architecture, avoiding the need for separate specialized systems for each detection type and reducing overall system complexity.
4Speed
If real-time processing of camera feeds is implemented, then responsiveness is improved, but computational load and energy consumption increase
Solution Approach 1:
The system segments processing tasks by resolution and complexity: the mobile robot performs lightweight real-time processing for immediate responsiveness to motion or obvious hazards, while the server handles more computationally intensive analysis. This segmentation enables real-time responsiveness without requiring the mobile device to consume excessive energy for all processing tasks.
Data Source
AI summary
Existing works on surveillance robots in indoor scenarios focus only on specific event detections and fail to detect most indoor objects due to lack of proper training. The present disclosure receives and pre-processes stream of input images specific to plurality of scenes related to indoor space from robot mounted camera. Pre-processed stream of input images is passed to a first trained model to obtain plurality of bounding boxes and masks pertaining to objects. An object property detector algorithm is run on masks to detect properties of objects. Classify signages comprised in masks as text signage by passing through optical character recognition or symbol by passing through second trained model. Convert output of optical character recognition and second trained model into facts and infer new facts using steam reasoning. Matching patterns comprised in form of an adaptable business logic with inferred new facts and trigger alerts, if pattern is matched.