Professional Documents
Culture Documents
Introduction Although the 1 Euro filter significantly optimizes tailing, jitter, and Experiments and Results
latency in interactions, this pre-machine learning era algorithm
Web browsers have become an indispensable tool for can be substantially empowered today by incorporating some We conducted two experiments, focusing on Mediapipe
human-computer interactions across various platforms. neural network-based temporal trajectory prediction algorithms. Hands across various devices. The standard for
In temporal trajectory prediction, LSTM can be employed to comparison was set using the GPU-accelerated Python
However, interactive performance (e.g the immediacy of
implement a ’Seq2Seq’ approach, where we use past trajectory version of Mediapipe. We compared the errors by
interaction, the accuracy of interaction data) in browser- data to predict future trajectory data. We can certainly build calculating the Mean Squared Error (MSE) of each key
based apps can be hindered by hardware limitations and an ’N to1’ prediction model, which predicts the next coordinate node, post-'FLOW' algorithm processing, against the
position based on the previous N steps, but such a sliding standard answer. It was observed that after optimization
network delays, resulting in unwanted jitter and input signal
window prediction method is too time-consuming under low with the FLOW algorithm, the accuracy and performance
delays that negatively impact the user experience. To tackle computational power. We recommend web developers to turn of the browser-based Mediapipe Hands improved
these challenges, we propose a novel approach that the 1 euro filter for reducing significant jitter, and then perform significantly compared to before the optimization.
latency statistics on the filtered data. For example, under a fixed
leverages effective filtering algorithms and lightweight sampling rate, we observe how many frames the filtered data
machine learning models. By integrating the 1 euro filter and lags behind the ground truth. We found that using the maximum
lag value multiplied by 1.5 or 2 as input to the LSTM yields
Long Short-Term Memory (LSTM) algorithm, we successfully
significantly improved prediction accuracy.
optimize the accuracy of Mediapipe-based gesture For the challenge of determining an uncertain frame delay, we
interactions and tangible table application within web innovatively use the alpha parameter in the filter to adaptively
select the predictive frames generated by the LSTM. By
browsers. Our findings demonstrate that this combined
applying a linear mapping on the prediction sequence using the
approach not only enhances input signal accuracy but also alpha parameter, this approach effectively addresses some
ensures seamless and highly responsive web browser overfitting issues that arise when using a fixed frame as the
predictive frame.
interactions.
The Mediapipe hand landmark Fast movements result in pronounced According to the theory of Jota et al. , we categorized the
detection in the browser faces on-screen smearing in TUIO based collected data into three stages according to the
inaccuracy in many scenarios tangible interaction Web App. The acceleration of the token in each frame. 1:initial reaction,
rendered position could not catch the Paradigm for Data Preprocessing and Training in FLOW 2:large ballistic motion, 3:feedback adjusted final
token at high speed. adjustments. Based on these three criteria, we have
organized the data into the following table: original data,
data after 1 Euro filter processing, data after prediction
using LSTM + mean latency frame, and data after 𝛼-
Methods mapped frame and subsequent filtering