3D Video Motion Detection for Filtering Nuisance Surveillance Events
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing surveillance systems face challenges in efficiently detecting relevant motion from nuisance events, such as tree motion, shadow, reflections, rain/snow, and flags, leading to numerous irrelevant motion events and high computational costs due to extensive processing requirements.
Innovation Solution
A unified, end-to-end data-driven method using Spatial-temporal Attention-based 3D Convolutional Neural Networks (3D ConvNets) that leverage spatial-temporal redundancies to efficiently detect relevant motion by parsing an entire video clip in one forward pass, incorporating background subtraction, object detection, and video tracking, with a compact and lightweight model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional object detection-based methods are used to detect relevant motion, then detection precision can be maintained, but processing speed decreases and computational costs increase significantly
Solution Approach 1:
The video processing task is segmented into two distinct phases: a training phase using traditional object detection methods to learn motion patterns, and an inference phase using a lightweight neural network for rapid detection. This segmentation allows the system to maintain high detection precision during training while achieving fast processing speeds during actual motion detection, resolving the contradiction between precision and speed
Solution Approach 2:
The system performs preliminary action by pre-training a neural network model using extensive object detection data during an offline training phase. This preliminary training enables the model to learn and store motion patterns in advance, so that during runtime, the system can perform rapid inference without executing computationally expensive object detection algorithms, thus achieving both high precision and fast processing speed
2Measurement precision
If extensive processing is performed to detect relevant motion from nuisance events, then detection accuracy improves, but computational costs increase
Solution Approach 1:
The invention extracts and isolates the essential motion detection function from the complex object detection pipeline. By using a simplified neural network that only needs to detect motion patterns rather than identify objects, the system achieves high detection accuracy while dramatically reducing computational costs and energy consumption, effectively extracting the core functionality needed for the application
3Reliability
If complex object detection algorithms are used, then relevant motion can be accurately identified, but processing time increases
Solution Approach 1:
The system creates a simplified copy of the motion detection capability by training a lightweight neural network to replicate the motion pattern recognition abilities of complex object detection algorithms. This copied model performs only the essential motion detection function without the overhead of full object detection, achieving reliable motion identification with minimal processing time
Data Source
AI summary
Methods, systems, and/or apparatuses are described for detecting relevant motion of objects of interest (e.g., persons and vehicles) in surveillance videos. As described herein input data based on a plurality of captured images and/or video is received. The input data may then be pre-processed and used as an input into a convolution network that may, in some instances, have elements that perform both spatial-wise max pooling and temporal-wise max pooling. The convolution network may be used to generate a plurality of prediction results of relevant motion of the objects of interest.