Live Video Ad Detection Using RGB Sequences and Neural Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting live-streamed advertisements on electronic devices with limited memory capacity are computationally intensive, making them difficult to implement effectively on devices like TVs, leading to user inconvenience.
Innovation Solution
An electronic device uses RGB value analysis and a neural network model to identify advertisement sequences in live-streamed videos, determining their presence and providing data such as end times without requiring extensive computation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If video signature method is used to detect advertisement videos, then advertisement detection accuracy is improved, but computation complexity increases making it difficult to apply on devices with limited memory capacity
Solution Approach 1:
The patent segments the video processing task into two distinct phases: an offline training phase where a neural network model is trained using video signature data, and an online detection phase where the trained model performs lightweight inference. This segmentation allows the computationally intensive work to be done once during training, while the actual advertisement detection during live streaming requires minimal computation, making it suitable for devices with limited memory capacity.
Solution Approach 2:
The patent performs preliminary action by training the neural network model in advance using a dataset of video signatures. The trained model is then stored and reused for multiple advertisement detection tasks. This preliminary training phase prepares the system so that during actual live streaming, only lightweight inference is needed, avoiding the need to perform complex video signature analysis in real-time on resource-constrained devices.
2Ease of operation
If real-time advertisement detection is implemented during live streaming, then user convenience is improved, but computation resources are excessively consumed
Solution Approach 1:
The patent creates a simplified copy of the advertisement detection capability through the trained neural network model. Instead of performing full video signature analysis on every live-streamed video, the system uses the trained model which has learned the essential patterns during offline training. This copied knowledge allows for fast, low-resource inference that maintains advertisement detection functionality while consuming minimal computation resources during live streaming.
3Measurement precision
If video signature analysis is performed on live-streamed videos, then advertisement identification capability is improved, but processing time increases
Solution Approach 1:
The patent performs the computationally intensive video signature analysis and pattern learning in advance during the offline training phase. The neural network model is trained using numerous video samples and their corresponding signatures, building a comprehensive understanding of advertisement patterns. During live streaming, the system only needs to perform fast inference using the pre-trained model, dramatically reducing processing time while maintaining high advertisement identification capability.
Data Source
AI summary
An electronic device including: a communication interface; a memory to store at least one instruction; and a processor, wherein the processor is configured to receive a first video including a plurality of frames from an external device through the communication interface, acquire reference data corresponding to an advertisement video included in the first video based on a first sequence acquired based on red, green, and blue (RGB) values of the plurality of frames included in the first video, and compare a second sequence corresponding to a second video received from the external device with the reference data to determine whether the second video is the advertisement video.