Mobile Robot Anomaly Detection Using RoI Motion Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomalous activity detection systems for mobile surveillance robots, such as robots and drones, face challenges due to their non-static camera positions, leading to inefficiencies in distinguishing unusual human activities from regular movements, and existing methods often rely on static camera assumptions, datasets prone to overfitting, and lack real-world applicability.
Innovation Solution
A spatio-temporal neural network architecture mimicking the Konio-Parvocellular-Magno cells of the human brain is used to detect unusual human activities by identifying a region of interest (RoI) and employing optical flow and Structural Similarity Index Measure (SSIM) to filter out static background motion, followed by a trained spatio-temporal neural network to classify activities as normal or anomalous.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If a mobile surveillance robot with non-static camera is used to patrol and detect anomalies, then the coverage area and dynamic monitoring capability are improved, but the difficulty of distinguishing unusual human activities from regular movements increases due to camera motion
Solution Approach 1:
The patent segments the video stream into multiple sub-sequences based on camera motion states. By dividing the continuous video feed into segments corresponding to different robot movement phases, the system can apply appropriate analysis methods to each segment, effectively separating the camera-induced motion from actual human activities in the scene.
Solution Approach 2:
The patent introduces an intermediary processing layer that analyzes camera motion patterns and uses this information to guide the anomaly detection process. This intermediary system acts as a mediator between the raw video input and the anomaly detection algorithm, compensating for camera motion effects before anomaly detection is performed.
2Device complexity
If existing anomaly detection algorithms designed for static cameras are applied to mobile surveillance robots, then the implementation complexity is reduced, but the detection accuracy deteriorates due to inability to handle camera motion
Solution Approach 1:
The patent transforms the static camera assumption into a dynamic model that explicitly accounts for camera motion. The system continuously adapts its analysis based on the robot's movement state, making the detection algorithm dynamic rather than static. This allows the system to maintain high detection accuracy while working with mobile platforms.
Solution Approach 2:
The patent changes the parameters used for anomaly detection based on camera motion characteristics. By adjusting detection thresholds, analysis windows, and processing parameters according to the robot's movement state, the system optimizes detection accuracy for each operational condition without requiring completely different algorithms.
3Ease of manufacture
If synthetic datasets are used for training anomaly detection models, then the training process is simplified and faster, but the model generalization capability and real-world applicability deteriorate
Solution Approach 1:
The patent performs preliminary actions to collect and annotate real-world video data from mobile surveillance robots in actual deployment environments. By gathering authentic training data before model development, the system ensures that the trained models are exposed to real-world variations and complexities, improving their generalization capability while maintaining practical training workflows.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Technical challenge in unusual human activity detection task is to rightly identify only unexpected or unusual movements from constant regular movements present in a scene, with most techniques built on understanding that camera is static. However, ego view camera of mobile surveillance robot is in motion as robot navigates. Embodiments herein provide a method and system for anomalous activity detection for mobile surveillance robots by mimicking 'Konio-Parvocellular-Magno' cells of the human brain into a NN model, which are responsible for detecting slow, normal, and swift changes in perceived scenes. To detect anomalous activity, the static or normal movements of scene captured by ego view camera are identified as redundant information and only RoI is forwarded for further processing using the Optical flow and SSIM techniques. The NN model mimicking KPM is trained only on the RoI to detect normal or anomalous activity.