Transfer Learning for Stable Video Analytics Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional video analytics systems exhibit significant fluctuations in accuracy and consistency across consecutive video frames, even in static scenes, due to dynamic camera parameter changes that affect pixel values, leading to inaccurate object and face detection.
Innovation Solution
Retrain deep learning models using transfer learning techniques, applying knowledge from image analytics tasks to video analytics tasks, and iteratively optimize model predictions to reduce fluctuations in accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional deep learning models trained on images are applied to video analytics, then object detection capability is achieved, but accuracy fluctuates significantly across consecutive video frames
Solution Approach 1:
The patent applies transfer learning to retrain the deep learning model using video frames from multiple scenarios, which changes the training parameters and data distribution. This adaptation allows the model to account for dynamic camera parameters and environmental factors, thereby reducing accuracy fluctuations across consecutive video frames while maintaining high detection accuracy
Solution Approach 2:
The patent performs preliminary retraining of the deep learning model using a diverse set of video frames from various scenarios before deploying it for actual video analytics. This preliminary action prepares the model to handle variations in camera settings and environmental conditions, preventing accuracy fluctuations before they occur during real-time processing
2Measurement precision
If deep learning models are retrained using transfer learning, then accuracy fluctuations are reduced, but processing time and computational resources increase
Solution Approach 1:
The patent applies transfer learning by leveraging pre-trained deep learning models and retraining them partially on video-specific data rather than training from scratch. This partial action approach achieves the necessary adaptation to reduce accuracy fluctuations while significantly reducing the time and computational resources required compared to complete retraining
3Measurement precision
If video analytics systems account for camera settings and environmental factors, then detection accuracy improves, but system complexity increases
Solution Approach 1:
The patent uses transfer learning as an intermediary mechanism that indirectly accounts for camera settings and environmental factors. Instead of explicitly modeling each factor, the retrained deep learning model learns to adapt to these variations through exposure to diverse training data, thereby improving detection accuracy without significantly increasing system complexity
Data Source
AI summary
Systems and methods are provided for increasing accuracy of video analytics tasks in real-time by acquiring a video, and identifying fluctuations in the accuracy of video analytics applications across consecutive frames of the video. The systems and methods can support decision-making based on results of video analytics tasks. The fluctuations are quantified based on an average relative difference of true-positive detection counts across consecutive frames. Fluctuations in accuracy are reduced by applying transfer learning to a deep learning model trained using images, and retraining the deep learning model using video frames. A quality of object detections is determined based on an amount of track-ids assigned by a tracker across different video frames. Optimization of the reduction of fluctuations includes iteratively repeating the identifying, the quantifying, the reducing, and the determining the quality of object detections until a threshold is reached. Frame predictions are generated using the retrained deep learning model.