Towards Safe Self-Stimulatory Behaviors in Autistic Children:
HarmAlert4AutisticChildren (HA4AC)
Aleenah Khan
a
and Hassan Foroosh
Department of Computer Science, University of Central Florida, Orlando, U.S.A.
Keywords:
Autism, Autism Spectrum Disorder, ASD, Self-Stimulatory Behaviors, Stimming, Self-Injurious Behaviors,
Stereotypical Behaviors, Self-Harm.
Abstract:
Self-Stimulatory behaviors, or stimming is quite common in Autism and can begin as early as infancy. Autistic
infants may show early signs of stimming through repetitive movements such as hand flapping, rocking, or
head banging. These stereotypical behaviors help self-regulation and are generally not harmful unless they
pose a safety risk (e.g., head banging) or significantly interfere with daily activities. In such cases, the parent
or caregiver must immediately intervene to ensure the safety of the child. To foster a safe environment for
autistic children, we introduce a novel problem of identifying potentially harmful self-stimulatory behaviors
to alert the parent / caregiver. To pave the way for research, we consolidated a video-based dataset “Har-
mAlert4AutisticChildren” which categorizes autism-related stimming behaviors into two categories: helpful
and harmful. We utilize existing publicly available video datasets that focus on a different problem of self-
stimulatory behavior classification in autism. The curation process is based on a systematic review of the
literature of clinical research studies that analyze the impacts of various self-stimulatory behaviors in autistic
children. In addition to introducing a new research problem and a new dataset, we also provide baseline re-
sults using the Contrastive Language-Image Pretraining (CLIP) model. The dataset and code are available on
GitHub: https://github.com/AleenahK/HarmAlert4AutisticChildren-HA4AC.
1 INTRODUCTION
According to the National Institute of Mental Health,
Autism Spectrum Disorder (ASD) is a lifelong neu-
rological and developmental disorder that can cause
significant social and behavioral challenges. Diagnos-
tic criteria for ASD involve the evaluation of social-
communication skills, including poor eye contact, dif-
ficulty maintaining conversations, and lack of devel-
opmentally appropriate peer relationships, in addi-
tion to the presence of restricted or repetitive behav-
iors such as stereotyped motor movements, hypo- or
hyper-sensitivities, and unusual interests (American
Psychiatric Association et al., 2013). It is known as
a spectrum disorder because there is a wide variation
in the type and severity of symptoms people experi-
ence.
Restricted and repetitive behaviors (RRBs) are
a core characteristic of ASD and are also referred
to as ’stereotyped behaviors’, ’stereotypy’, ’self-
stimulatory behaviors’ or ’stimming’ in clinical lit-
erature. We will use these terms interchangeably
a
https://orcid.org/0009-0001-2807-1551
throughout the remainder of the article. These be-
haviors include a range of actions including but not
limited to hand flapping, head banging, finger tap-
ping, spinning, scratching, clapping, rocking back
and forth, and lining up and flapping objects. It
also includes producing auditory stimuli, such as
whistling, humming, or idiosyncratic speech.
Early research studies declared these stereotypi-
cal behaviors redundant and discussed how these can
negatively impact autistic people by causing myriad
difficulties such as hindrance in learning capabilities
(Koegel and Covert, 1972), and social interactions
(Koegel et al., 1974). It is important to note that
these initial studies were based on very small groups
of autistic children.
Recently, there has been an increase in research
that attempts to shift the focus towards exploring the
experiences of autistic adults. Based on thematic
analysis of qualitative data obtained through question-
naires, interviews, and focus groups of autistic adults,
the researchers aim to understand their experiences
and perceptions of self-stimulatory behaviors. As a
result, it is revealed that self-stimulatory behaviors
986
Khan, A. and Foroosh, H.
Towards Safe Self-Stimulatory Behaviors in Autistic Children: HarmAlert4AutisticChildren (HA4AC).
DOI: 10.5220/0013389700003912
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 20th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2025) - Volume 2: VISAPP, pages
986-994
ISBN: 978-989-758-728-3; ISSN: 2184-4321
Proceedings Copyright © 2025 by SCITEPRESS – Science and Technology Publications, Lda.

Figure 1: Summarizing Clinical Studies on ASD-related Self-Injurious Self-Stimulatory Behaviors.
help regulate intense emotions, dissipate anxiety, and
manage sensory sensitivities, while suppressing these
leads to further stress (Kapp et al., 2019), (Charlton
et al., 2021).
According to the National Autistic Society, self-
stimulatory behaviors are often very helpful and en-
joyable; however, some of these can be self-injurious,
for example, head-banging, scratching, biting, skin
picking, hair pulling, etc.
According to a 2017 study (Guan and Li, 2017)
published in the American Journal of Public Health,
individuals with a diagnosis of autism are at a sub-
stantially heightened risk of death due to injury. Dur-
ing the 16-year study period, about 27.9 % of the
deaths in autistic individuals were attributed to injury
mortality. In addition, deaths due to unintentional
injury in autistic individuals were nearly 3 times as
likely as in the general population especially for chil-
dren under 15 years.
Hence, we conclude that self-stimulatory behav-
iors can be either helpful or harmful. It is crucial
for the parent / caregiver to analyze the situation and
act accordingly. It is very important to provide a
supportive environment to allow autistic people to
freely engage in stimming without having to suppress
it. On the contrary, if the behavior is causing any
kind of harm like self-injurious behaviors, the par-
ent/caregiver should address it with appropriate inter-
ventions.
We realize that ASD not only has significant neg-
ative impacts on a child’s development, but it also af-
fects their family’s social, emotional, and economic
well-being. Providing care and support for autistic
people requires a significant investment of time and
effort. Parents and caregivers frequently report ex-
periencing stress and anxiety related to caring for an
autistic child. (Estes et al., 2013), (Lecavalier et al.,
2006).
In an effort to help autistic people and their fam-
ilies, we propose a new computer vision task with
the objective of developing models that can distin-
guish between helpful and harmful self-stimulatory
behaviors to send alerts for intervention. The very
initial step towards building such an automated sys-
tem is to provide a standard, publicly available, video-
based dataset representing both helpful and poten-
tially harmful behaviors. This research paper focuses
on utilizing existing ASD-related video datasets, orig-
inally formulated for other tasks, to curate a new
dataset.
2 DATASET CURATION
A high-quality video dataset aligned with helpful
and harmful self-stimulatory behaviors is crucial for
building robust automated systems that are capable
of identifying self-injurious behaviors and alerting the
parent or caregiver to intervene and ensure the safety
of autistic children.
Towards Safe Self-Stimulatory Behaviors in Autistic Children: HarmAlert4AutisticChildren (HA4AC)
987

Figure 2: Harm Alert 4 Autistic Children - HA4AC Dataset.
Unfortunately, to the best of our knowledge, such
a dataset does not exist, and due to ethical concerns,
like identity protection, creating it from scratch is
also challenging. However, the computer vision re-
search community has been working towards building
datasets that represent self-stimulatory behaviors with
the aim of developing diagnostic systems to iden-
tify early signs of autism. Early diagnosis and in-
tervention can significantly improve verbal and non-
verbal communication, learning capabilities, social
reciprocity, and overall well-being of autistic chil-
dren. These datasets have been collected from pub-
licly available videos of autistic people filmed by their
parents or caregivers and shared on social media plat-
forms such as YouTube.
These ASD-related datasets have representations
of different self-stimulatory behaviors, such as, head
banging and hand flapping, however they don’t clas-
sify the nature of the behavior as positive or negative.
We aim to leverage these publicly available datasets
to curate a new dataset by categorizing the available
action classes as either HELPFUL or HARMFUL.
To avoid any kind of personal bias, instead of re-
lying on our instincts, we perform a systematic liter-
ature review of behavioral studies conducted by clin-
ical researchers to analyze both positive and negative
impacts of various self-stimulatory behaviors.
Based on this analysis, we then bifurcate the
self-stimulatory behaviors present in the existing
ASD-related datasets and propose the new ”Har-
mAlert4AutisticChildren” dataset to identify poten-
tially problematic stimming behaviors in autistic chil-
dren.
In the following section, we present a system-
atic review of clinical research studies focused on
autism-related self-stimulatory behaviors to analyze
their positive and negative impact on autistic people
to classify them as helpful or harmful.
2.1 Clinical Studies
To understand the impacts of different self-
stimulatory behaviors, such as, their functions
(e.g. self-regulation, sensory stimulation) or their
consequences (e.g. self-harm), we conduct a sys-
tematic study of existing clinical literature. We
aim to distinguish between helpful or harmless
self-stimulatory behaviors, especially those that can
potentially cause harm by identifying the overlap
between self-stimulatory and self-injurious behaviors
in autistic people.
2.1.1 Search Keywords
We identify four sets of keywords to search for rel-
evant behavioral studies related to autism, namely:
Context, Neutral, Positive, and Negative. Each of
these sets represents field-specific jargon and is listed
as follows.
• Context: self-stimulation, self-stimulatory be-
haviors, stereotypical behaviors, stereotypy, stim-
ming, autism, ASD, fidgeting
• Neutral: statistics, impacts, affects, analysis, in-
sights, prevalence, frequency
• Positive: helpful, benefits, positive, self-soothing,
emotional regulation, healthy, de-stress
VISAPP 2025 - 20th International Conference on Computer Vision Theory and Applications
988

• Negative: risk factors, issues, difficulties, nega-
tive, harm, injury, self-injurious, self-harm, dan-
gerous, aggressive
2.1.2 Search Queries
We formulate search queries using the appropriate
combinations of search keywords mentioned in Sec-
tion 2.1.1. To retrieve relevant articles, we use both
generic and action-specific queries. Action-specific
queries include additional keywords that represent
self-stimulatory behaviors such as ”head banging”
and ”hand flapping”. We provide one example of each
of these categories in neutral, positive, and negative
contexts, respectively.
• Generic Queries
– Impacts of self-stimulatory behaviors in
Autism
– Autistic self-stimulation and emotional regula-
tion
– Self-injurious stereotypical behaviors
• Action-Specific Queries
– Hyper-sensitivities in autistic individuals
– Helpful vocal stims
– Head-banging injuries in autism
2.1.3 Search Results & Analysis
In this section, we discuss the results of our search
and provide analysis that helped us categorize self-
stimulatory behaviors as helpful and harmful. We first
share insights about the helpful stimming behaviors
followed by those that are potentially harmful accord-
ing to evidences from clinical studies.
Studies that highlight the benefits of self-
stimulatory behaviors and advocate for them are
based on first-person accounts of autistic adults.
Autistic children often struggle to communicate and
express their feelings due to the prevalence of non-
verbality and minimum verbality. Due to this reason,
they are also not able to explain the significance of
self-stimulatory behaviors in their life. With the help
of early intervention therapies, autistic children man-
age to gain language abilities later in their life.
Research studies based on first-person accounts of
autistic adults advocate in favor of self-stimulatory
behaviors and describe stimming using words with
deeply positive connotations such as ’calming’, ’com-
forting’, ’soothing’, ’joyful’ and ’enjoyable’. Stim-
ming helps autistic people overcome feelings of ner-
vousness, anxiety, and anger, or express happiness
and excitement (Kapp et al., 2019). It also helps autis-
tic people organize their thoughts, improve focus, and
get rid of excessive energy (Joyce et al., 2017).
We present a summarized list of different motor,
vocal, and visual stereotypes that often help autistic
people based on personal experiences shared through
questionnaires, focus groups, and questionnaires con-
ducted in these behavioral studies.
• Motor Stereotypes: hand flapping, body rock-
ing, pacing back and forth, finger tapping, spin-
ning, twirling pen or jewelry, doodling, jumping
or bouncing
• Vocal Stereotypes: humming, whistling,
echolalia, use of atypical language
• Visual Inspection: aligning objects, spinning ob-
jects
We also share some personal accounts of autis-
tic individuals in their own words which convey their
sentiments in a persuasive way.
• ”People should be allowed to do what they like” -
(Kapp et al., 2019)
• ”Stim your heart out” & ”Syndrome rebel” -
(Stevenson, 2020)
• “It feels like holding back something you need to
say” - (Charlton et al., 2021)
• ”If I don’t Do It, I’m Out of Rhythm and I Can’t
Focus As Well” - (McCormack et al., 2023)
• ”I Wish They’d Just Let Us Be” - (Sagar et al.,
2023)
• ”It Helps Make the Fuzzy Go Away” - (Friedman
et al., 2024)
Next, we discuss clinical studies that report the
prevalence of negative or harmful stereotypical be-
haviors related to autism. Self-harm or self-injurious
behavior (SIB) is a major concern for autistic chil-
dren and adolescents. These behaviors are defined
as non-accidental, non-suicidal, self-inflicted actions
that result in physical injury (Yates, 2004). Examples
of such behaviors include self-biting, self-hitting, hair
pulling, skin picking, scratching, etc. (Furniss and
Biswas, 2012), (Guan and Li, 2017), (Maddox et al.,
2017). We summarize the results of our search for
harmful self-stimulatory behaviors in Figure 1. With
the help of an Upset Plot (Lex et al., 2014), we visu-
alize the intersection of different stereotypical actions
studied in these research studies. The latter half of
the figure represents the occurrences of these behav-
iors in clinical literature (left) and the different combi-
nations studied together (right), while the former half
provides references of these studies. Self-hitting, self-
biting, self-scratching, and pulling hair are the most
reported stimming behaviors with respect to Autism
Spectrum Disorder.
Towards Safe Self-Stimulatory Behaviors in Autistic Children: HarmAlert4AutisticChildren (HA4AC)
989

Figure 3: Examples of Helpful & Harmful Self-Stimulatory Behaviors.
Table 1: Comparison of Existing ASD-related Datasets used in the “HarmAlert4Autistic” Dataset.
SSBD ESBD SSBD+ ASBD AV-ASD
Original Size 75 141 61 165 928
Categories 3 4 3 4 10
Self- ArmFlapping ArmFlapping ArmFlapping ArmFlapping AbsenceOrAvoidanceOfEyeContact
Stimulatory HeadBanging HandAction HeadBanging HandAction AggressiveBehavior
Behaviors Spinning HeadBanging Spinning HeadBanging Background
Spinning Spinning HyperOrHyporeactivityToSensoryInput
Non-ResponsivenessToVerbalInteraction
Non-TypicalLanguage
ObjectLining-Up
Self-HittingOrSelf-InjuriousBehavior
Self-SpinningOrSpinningObjects
UpperLimbStereotypies
Annotated Yes No Yes Yes Yes
Source YouTube YouTube YouTube YouTube YouTube
Vimeo Facebook
DailyMotion
Release Year 2013 2021 2023 2023 2024
2.2 Self-Stimulatory Datasets
In this section, we study existing ASD-related video
datasets that have been the focus of the computer vi-
sion research community in the past decade. We dis-
cuss these datasets in chronological order of their ex-
istence. A summarized comparison of these datasets
is presented in Table 1.
2.2.1 Self-Stimulatory Behavior Dataset (SSBD)
The first attempt to create a video dataset for model-
ing self-stimulatory behaviors related to Autism Spec-
trum Disorder was made by (Rajagopalan et al., 2013)
in 2013. The Self-Stimulatory Behavior Dataset
(SSBD) consisted of 75 distinct videos distributed
into three categories: Arm Flapping, Head Banging,
and Spinning. These videos were recorded in natu-
ral settings by parents or caregivers of autistic chil-
dren and were collected from the popular social media
website YouTube.
2.2.2 Expanded Stereotype Behavior Dataset
(ESBD)
In 2021, (Negin et al., 2021) proposed a larger dataset
called the Expanded Stereotype Behavior Dataset
(ESBD). The new dataset consisted of a total of 141
videos, approximately twice the size of the SSBD
dataset. They also added a new class label referred
as ”Hand Action” together with the existing classes,
Arm Flapping, Head Banging, and Spinning. The
problem with the ESBD dataset is that it was not prop-
erly annotated with start time and end time of the
stimming behaviors. We provide proper annotations
for all videos in the ESBD dataset to include them in
our new dataset.
VISAPP 2025 - 20th International Conference on Computer Vision Theory and Applications
990

2.2.3 Updated Self-Stimulatory Behavior
Dataset (SSBD+)
(Wei et al., 2023) made an effort to expand the ex-
isting SSBD dataset by including 12 new videos and
removing 11 noisy videos. Like the SSBD dataset, the
new videos were also added from YouTube. The final
dataset consisted of 61 unique and noise-free videos
spanning the same three categories: Arm Flapping,
Head Banging, and Spinning.
2.2.4 Autism Stimming Behavior Dataset
(ASBD)
Recently, (Ribeiro et al., 2023) combined all the
aforementioned datasets: SSBD, ESBD and Updated
SSBD to create a new consolidated dataset called
the Autism Stimming Behavior Dataset (ASBD). The
final dataset consisted of 165 distinct videos that
spanned four classes. They also provide annotations
for start time and duration of the stimming actions.
2.2.5 Audio-Visual Autism Spectrum Dataset
(AV-ASD)
(Deng et al., 2024) recently introduced a more ex-
tensive audio-visual dataset to stimulate further re-
search for the diagnosis of autism-related behaviors.
Unlike preceding datasets, AV-ASD includes both so-
cial interaction challenges, and restricted and repeti-
tive behaviors (RRBs). The AV-ASD dataset is also
richer in terms of both quality and quantity, having
a much greater number of both categories and sam-
ples. This dataset comprises 928 clips extracted from
569 unique videos distributed in 10 categories. These
video clips include diverse behaviors and environ-
ment settings and are collected from popular social
media platforms YouTube and Facebook. The dataset
provides multiple labels for each video clip consider-
ing the fact that an autistic individual can exhibit mul-
tiple autistic behaviors at the same time. The dataset
also provides precise time-stamp annotations for the
start and end of each autistic behavior.
3 HarmAlert4AutisticChildren
The HarmAlert4AutisticChildren (HA4AC) dataset is
our effort to consolidate a new dataset using videos
from the five existing datasets discussed in the previ-
ous section. It is the first dataset designed to enable
research towards the development of automated sys-
tems that should be capable to distinguish between
helpful/harmless and harmful self-stimulatory behav-
iors. The proposed dataset consists of a total of
731 video clips extracted from 368 distinct videos
downloaded from various social media apps, such as
YouTube, Facebook, Vimeo and DailyMotion. These
videos are captured in realistic, unbounded scenes and
include diverse behaviors which fall in one of the two
categories: Helpful (527 clips; 10761 s) or Harmful
(204 clips; 3437 s). The shortest clip has a duration
of 2 seconds, while the longest clip is 14 minutes and
48 seconds long. All video clips have carefully as-
signed time stamp annotations representing start time
and end time of the self-stimulatory behavior exhib-
ited in the clip.
Based on the analysis presented in Section 2.1.3
and Figure 1 of this article, we are able to identify
the following self-stimulatory behaviors from exist-
ing datasets as ’Harmful’: Head Banging, Aggres-
sive Behaviors, and Self-Hitting Or Self-Injurious
Behaviors. The rest of the self-stimulatory behav-
iors are considered as ’Helpful’ as they seem to be
harmless considering the literature. However, these
stimming behaviors do not always occur in isolation
and often overlap with each other. Figures 2 (a) and
2 (b) represent the percentage of time duration and
frequency of occurrence of these stereotypical behav-
iors that show their contribution to both Helpful and
Harmful classes, respectively. Figure 2 (c) represents
the gender-wise distribution of the data indicating the
dominance of male autistic children in the dataset.
This disparity aligns with research studies that claim
autism spectrum disorder is more commonly diag-
nosed in males than females. The male-to-female ra-
tio in autism diagnoses is often cited as around 4:1.
According to another research study (Schuck et al.,
2019), females camouflage the symptoms of ASD
more than males potentially contributing to the dif-
ference in prevalence.
Figure 3 presents a set of five samples for each of
the two class categories.
4 EXPERIMENTS
For benchmarking purposes, we evaluate the perfor-
mance of CLIP-based models in zero-shot settings for
our newly curated HA4AL dataset. Below are the ex-
perimental details of the baseline models.
4.1 Contrastive Language-Image
Pretraining (CLIP)
The Contrastive Language-Image Pretraining (CLIP)
model (Radford et al., 2021) is a multi-modal vi-
sion and language model which maps image and text
pairs to a shared embedding space. CLIP is widely
Towards Safe Self-Stimulatory Behaviors in Autistic Children: HarmAlert4AutisticChildren (HA4AC)
991

known for its ability to generalize and perform zero-
shot learning effectively. Despite originally being
designed for images, CLIP can be easily adapted to
work with videos. We use the following two strate-
gies to evaluate the CLIP model on our downstream
task.
4.1.1 Vanilla CLIP
The most straightforward approach to adapting the
CLIP model (Radford et al., 2021) for video classifi-
cation is to apply temporal pooling to the embeddings
of individual frames, thus generating a unified repre-
sentation. In our experiments, we used the ViT-B/16,
ViT-B/32, and ViT-L/14 models in zero-shot settings.
4.1.2 Video Fine-Tuned CLIP
The ViFi-CLIP model (Rasheed et al., 2023) em-
ploys video-based fine-tuning of the image-based
CLIP model to bridge the domain gap between im-
ages and videos. The input video frames are first pro-
cessed using the CLIP image encoder to obtain fea-
ture embeddings. These embeddings are then inte-
grated through feature pooling, followed by similar-
ity matching with the corresponding text embeddings.
The ViFi-CLIP ViT-B/32 model used in our experi-
ments is fine-tuned on the Kinetics-400 dataset (Kay
et al., 2017). Kinetics-400 is a human action dataset
with 400 classes and at least 400 video clips per
class covering a wide range of both human-action and
human-human interactions. The fine-tuning process
is performed for 10 epochs, and the resulting model
is evaluated on the downstream HA4AL dataset un-
der zero-shot settings.
4.2 Evaluation Metric
To evaluate the performance of the CLIP-based mod-
els in our downstream task using the HA4AL dataset,
we use accuracy as the evaluation metric.
Table 2: Experimental Results.
Model Accuracy
Vanilla CLIP
ViT-B/16 37.2 %
ViT-B/32 41.9 %
ViT-L/14 56.1 %
ViFi-CLIP ViT-B/32 50.7 %
4.3 Results & Analysis
The zero-shot evaluations of both vanilla CLIP and
ViFi-CLIP on our downstream task show impressive
generalization capability of the CLIP model in Table
2. In case of Vanilla CLIP, the ViT-L/14 model per-
forms better than the other variants ViT-B/16 and ViT-
B/32 with a 56.1 % accuracy. The reason being that
the ViT-L14 model has a larger configuration with
more transformer layers and, therefore, more learn-
able parameters. The ViT-L/14 has a 14x14 patch
size which enables it to capture finer details and in-
creases it’s capacity to learn complex relationships
in the data. To compare Vanilla CLIP with ViFi-
CLIP, we use the same ViT-B/32 configuration. As
expected, ViFi-CLIP outperforms Vanilla CLIP with
an 8.8 % improvement due to the fine-tuning advan-
tage. It is unsurprising that the model with the largest
configuration, ViT-L/14 outperforms the rest.
5 CONCLUSION
In this research paper, we introduce a new computer
vision-based recognition task to identify potentially
harmful and harmless stereotypical behaviors in autis-
tic population. We took the first step towards solv-
ing this problem by proposing a new dataset, the Har-
mAlert4AutisticChildren (HA4AC) dataset. We per-
form a systematic review of the existing clinical lit-
erature to understand the topography, functions, and
consequences of self-stimulatory behaviors to catego-
rize them as helpful and harmful. By evaluating exist-
ing datasets for self-stimulatory behavior recognition,
we filter positive and negative examples of exhibiting
self-harm and aggression. We also present baseline
results using CLIP-based video classification models
to benchmark future research efforts.
6 FUTURE DIRECTIONS
The proposed HA4AC dataset suffers from a class
imbalance problem, with helpful stimming behaviors
constituting 74 % and harmful behaviors compris-
ing only 26 %, at a ratio of 3.55:1. Furthermore,
the HA4AC dataset exhibits gender imbalance, with
the male population comprising 74 % of the data
points and autistic female children underrepresented.
In our future work, we plan to work on this class
imbalance problem and employ state-of-work large-
language models to improve the accuracy on the video
classification task.
REFERENCES
Akram, B., Batool, M., Rafi, Z., and Akram, A. (2017).
Prevalence and predictors of non-suicidal self-injury
VISAPP 2025 - 20th International Conference on Computer Vision Theory and Applications
992

among children with autism spectrum disorder. Pak-
istan Journal of Medical Sciences, 33(5):1225.
American Psychiatric Association, D., American Psychi-
atric Association, D., et al. (2013). Diagnostic and
statistical manual of mental disorders: DSM-5, vol-
ume 5. American psychiatric association Washington,
DC.
Charlton, R. A., Entecott, T., Belova, E., and Nwaordu,
G. (2021). “it feels like holding back something you
need to say”: Autistic and non-autistic adults accounts
of sensory experiences and stimming. Research in
Autism Spectrum Disorders, 89:101864.
Deng, S., Kosloski, E. E., Patel, S., Barnett, Z. A., Nan, Y.,
Kaplan, A., Aarukapalli, S., Doan, W. T., Wang, M.,
Singh, H., et al. (2024). Hear me, see me, understand
me: Audio-visual autism behavior recognition. arXiv
preprint arXiv:2406.02554.
Dominick, K. C., Davis, N. O., Lainhart, J., Tager-Flusberg,
H., and Folstein, S. (2007). Atypical behaviors in
children with autism and children with a history of
language impairment. Research in developmental dis-
abilities, 28(2):145–162.
Duerden, E. G., Oatley, H. K., Mak-Fan, K. M., McGrath,
P. A., Taylor, M. J., Szatmari, P., and Roberts, S. W.
(2012). Risk factors associated with self-injurious be-
haviors in children and adolescents with autism spec-
trum disorders. Journal of autism and developmental
disorders, 42:2460–2470.
Estes, A., Olson, E., Sullivan, K., Greenson, J., Winter,
J., Dawson, G., and Munson, J. (2013). Parenting-
related stress and psychological distress in mothers of
toddlers with autism spectrum disorders. Brain and
Development, 35(2):133–138.
Folch, A., Cort
´
es, M., Salvador-Carulla, L., Vicens, P.,
Iraz
´
abal, M., Mu
˜
noz, S., Rovira, L., Orejuela, C.,
Haro, J., Vilella, E., et al. (2018). Risk factors and to-
pographies for self-injurious behaviour in a sample of
adults with intellectual developmental disorders. Jour-
nal of Intellectual Disability Research, 62(12):1018–
1029.
Friedman, S., Noble, R., Archer, S., Gibson, J., and Hughes,
C. (2024). “it helps make the fuzzy go away”: Autistic
adults’ perspectives on nature’s relationship with well-
being through the life course. Autism in Adulthood,
6(2):192–204.
Furniss, F. and Biswas, A. (2012). Recent research on
aetiology, development and phenomenology of self-
injurious behaviour in people with intellectual disabil-
ities: a systematic review and implications for treat-
ment. Journal of Intellectual Disability Research,
56(5):453–475.
Guan, J. and Li, G. (2017). Injury mortality in individu-
als with autism. American journal of public health,
107(5):791–793.
Gulsrud, A., Lin, C., Park, M., Hellemann, G., and Mc-
Cracken, J. (2018). Self-injurious behaviours in
children and adults with autism spectrum disorder
(asd). Journal of intellectual disability research,
62(12):1030–1042.
Handen, B. L., Mazefsky, C. A., Gabriels, R. L., Peder-
sen, K. A., Wallace, M., and Siegel, M. (2018). Risk
factors for self-injurious behavior in an inpatient psy-
chiatric sample of children with autism spectrum dis-
order: A naturalistic observation study. Journal of
autism and developmental disorders, 48:3678–3688.
Hattier, M. A., Matson, J. L., Belva, B. C., and Horovitz,
M. (2011). The occurrence of challenging behaviours
in children with autism spectrum disorders and atypi-
cal development. Developmental Neurorehabilitation,
14(4):221–229.
Joyce, C., Honey, E., Leekam, S. R., Barrett, S. L., and
Rodgers, J. (2017). Anxiety, intolerance of uncer-
tainty and restricted and repetitive behaviour: Insights
directly from young people with asd. Journal of
Autism and Developmental Disorders, 47:3789–3802.
Kamio, Y. (2002). Self-injurious and aggressive behavior
in adolescents with intellectual disabilities: A com-
parison of adolescents with and without autism. The
Japanese Journal of Special Education, 39(6):143–
154.
Kapp, S. K., Steward, R., Crane, L., Elliott, D., Elph-
ick, C., Pellicano, E., and Russell, G. (2019). ‘peo-
ple should be allowed to do what they like’: Autistic
adults’ views and experiences of stimming. Autism,
23(7):1782–1792.
Kay, W., Carreira, J., Simonyan, K., Zhang, B., Hillier, C.,
Vijayanarasimhan, S., Viola, F., Green, T., Back, T.,
Natsev, P., et al. (2017). The kinetics human action
video dataset. arXiv preprint arXiv:1705.06950.
Koegel, R. L. and Covert, A. (1972). The relationship
of self-stimulation to learning in autistic children 1.
Journal of Applied Behavior Analysis, 5(4):381–387.
Koegel, R. L., Firestone, P. B., Kramme, K. W., and Dunlap,
G. (1974). Increasing spontaneous play by suppress-
ing self-stimulation in autistic children 1. Journal of
Applied Behavior Analysis, 7(4):521–528.
Lecavalier, L. (2006). Behavioral and emotional problems
in young people with pervasive developmental disor-
ders: Relative prevalence, effects of subject character-
istics, and empirical classification. Journal of autism
and developmental disorders, 36:1101–1114.
Lecavalier, L., Leone, S., and Wiltz, J. (2006). The impact
of behaviour problems on caregiver stress in young
people with autism spectrum disorders. Journal of in-
tellectual disability research, 50(3):172–183.
Lex, A., Gehlenborg, N., Strobelt, H., Vuillemot, R., and
Pfister, H. (2014). Upset: visualization of intersecting
sets. IEEE transactions on visualization and computer
graphics, 20(12):1983–1992.
Licence, L., Oliver, C., Moss, J., and Richards, C. (2020).
Prevalence and risk-markers of self-harm in autistic
children and adults. Journal of autism and develop-
mental disorders, 50(10):3561–3574.
Maddox, B. B., Trubanova, A., and White, S. W. (2017).
Untended wounds: Non-suicidal self-injury in adults
with autism spectrum disorder. Autism, 21(4):412–
422.
Malhi, P. and Sankhyan, N. (2021). Intentional self harm
in children with autism. The Indian Journal of Pedi-
atrics, 88:158–160.
Towards Safe Self-Stimulatory Behaviors in Autistic Children: HarmAlert4AutisticChildren (HA4AC)
993

McCormack, L., Wong, S. W., and Campbell, L. E. (2023).
‘if i don’t do it, i’m out of rhythm and i can’t focus
as well’: Positive and negative adult interpretations of
therapies aimed at ‘fixing’their restricted and repeti-
tive behaviours in childhood. Journal of Autism and
Developmental Disorders, 53(9):3435–3448.
Moseley, R., Gregory, N. J., Smith, P., Allison, C., and
Baron-Cohen, S. (2019). A ‘choice’, an ‘addiction’,
a way ‘out of the lost’: exploring self-injury in autis-
tic people without intellectual disability. Molecular
autism, 10:1–23.
Negin, F., Ozyer, B., Agahian, S., Kacdioglu, S., and Ozyer,
G. T. (2021). Vision-assisted recognition of stereotype
behaviors for early diagnosis of autism spectrum dis-
orders. Neurocomputing, 446:145–155.
Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G.,
Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark,
J., et al. (2021). Learning transferable visual models
from natural language supervision. In International
conference on machine learning, pages 8748–8763.
PMLR.
Rajagopalan, S., Dhall, A., and Goecke, R. (2013). Self-
stimulatory behaviours in the wild for autism diagno-
sis. In Proceedings of the IEEE International Confer-
ence on Computer Vision Workshops, pages 755–761.
Rasheed, H., Khattak, M. U., Maaz, M., Khan, S., and
Khan, F. S. (2023). Fine-tuned clip models are effi-
cient video learners. In Proceedings of the IEEE/CVF
Conference on Computer Vision and Pattern Recogni-
tion, pages 6545–6554.
Ribeiro, G. O., Grellert, M., and Carvalho, J. T. (2023).
Stimming behavior dataset-unifying stereotype be-
havior dataset in the wild. In 2023 IEEE 36th In-
ternational Symposium on Computer-Based Medical
Systems (CBMS), pages 225–230. IEEE.
Richards, C., Davies, L., and Oliver, C. (2017). Predictors
of self-injurious behavior and self-restraint in autism
spectrum disorder: Towards a hypothesis of impaired
behavioral control. Journal of Autism and Develop-
mental Disorders, 47:701–713.
Richards, C., Moss, J., Nelson, L., and Oliver, C. (2016).
Persistence of self-injurious behaviour in autism spec-
trum disorder over 3 years: a prospective cohort study
of risk markers. Journal of neurodevelopmental dis-
orders, 8:1–12.
Richards, C., Oliver, C., Nelson, L., and Moss, J. (2012).
Self-injurious behaviour in individuals with autism
spectrum disorder and intellectual disability. Journal
of Intellectual Disability Research, 56(5):476–489.
Sagar, E., Khera, S. N., and Garg, N. (2023). “i wish
they’d just let us be.” experiences of indian autistic
individuals around stimming behaviors at the work-
place. Autism in Adulthood.
Schuck, R. K., Flores, R. E., and Fung, L. K. (2019). Brief
report: Sex/gender differences in symptomology and
camouflaging in adults with autism spectrum disor-
der. Journal of autism and developmental disorders,
49:2597–2604.
Slingsby, B., Yatchmink, Y., and Goldberg, A. (2017). Typ-
ical skin injuries in children with autism spectrum dis-
order. Clinical pediatrics, 56(10):942–946.
Stevenson, P. (2020). ’stim your heart out’and’syndrome
rebel’(performance artworks, autism advocacy and
mental health). idea journal, 17(2).
Wei, P., Ahmedt-Aristizabal, D., Gammulle, H., Denman,
S., and Armin, M. A. (2023). Vision-based activity
recognition in children with autism-related behaviors.
Heliyon, 9(6).
Yates, T. M. (2004). The developmental psychopathology
of self-injurious behavior: Compensatory regulation
in posttraumatic adaptation. Clinical psychology re-
view, 24(1):35–74.
VISAPP 2025 - 20th International Conference on Computer Vision Theory and Applications
994