added abstract

Author: anish-lakkapragada

README.md

@@ -5,7 +5,7 @@
 <p><strong><a href="mailto:anish.lakkapragada@gmail.com,peter100@stanford.edu,dpwall@stanford.edu">Authors</a></strong>: <a href="mailto:anish.lakkapragada@gmail.com">Anish Lakkapragada</a>, <a href="mailto:peter100@stanford.edu">Peter Washington</a>, <a href="mailto:dpwall@stanford.edu">Dennis P. Wall</a></p>
 <details>
    <summary>Abstract</summary>
-   <p> <em> A formal autism diagnosis can be an inefficient and lengthy process. Families may wait months orlonger before receiving a diagnosis for their child. One approach to lessen delays is the use digitaltechnologies to detect the presence of behaviors indicative of autism, which in aggregate may leadto remote and automated diagnostics. One of the strongest indicators of autism is stimming, whichincludes repetitive, self-stimulatory behaviors such as hand flapping, headbanging, and spinning.Using computer vision to detect hand flapping is especially difficult due to the sparsity of publictraining data in this space and excessive shakiness and motion in such data. Our work demonstratesa novel method that may overcome these issues:  we use hand landmark detection over time as afeature representation which is then fed into a Long Short-Term Memory (LSTM) model. We achieve avalidation accuracy and F1 Score of about 72% on detecting whether videos from the Self-StimulatoryBehaviour Dataset (SSBD) contain hand flapping or not. Our best model also predicts accurately onexternal videos we recorded of ourselves outside of the dataset it was trained on. This model usesless than 26,000 parameters, providing promise for fast deployment into ubiquitous and wearabledigital settings for a remote autism diagnosis </em> </p>
+   <p> <em> A formal autism diagnosis can be an inefficient and lengthy process. Families may wait months orlonger before receiving a diagnosis for their child. One approach to lessen delays is the use digitaltechnologies to detect the presence of behaviors indicative of autism, which in aggregate may leadto remote and automated diagnostics. One of the strongest indicators of autism is stimming, whichincludes repetitive, self-stimulatory behaviors such as hand flapping, headbanging, and spinning.Using computer vision to detect hand flapping is especially difficult due to the sparsity of publictraining data in this space and excessive shakiness and motion in such data. Our work demonstratesa novel method that may overcome these issues:  we use hand landmark detection over time as afeature representation which is then fed into a Long Short-Term Memory (LSTM) model. We achieve avalidation accuracy and F1 Score of about 72% on detecting whether videos from the Self-StimulatoryBehaviour Dataset (SSBD) contain hand flapping or not. Our best model also predicts accurately onexternal videos we recorded of ourselves outside of the dataset it was trained on. This model usesless than 26,000 parameters, providing promise for fast deployment into ubiquitous and wearabledigital settings for a remote autism diagnosis. </em> </p>
 </details>
 <h2 id="objective">Objective</h2>
 <p>Today&#39;s autism diagnosis is a quite lengthy and inefficient process. Families often have to wait a few years before receiving a diagnosis, and this problem is exacerbated by the fact that the earliest possible intervention is required for best clinical outcomes. One of the biggest indicators of autism is self-stimulatory, or stimming, behaviors such as hand flapping, headbanging, and spinning. In this paper, we demonstrate successful lightweight detection of hand flapping in videos using deep learning and activity recognition. We believe such methods will help create a remote, fast, and accessible autism diagnosis. </p>