Towards a Forensic Event Ontology to Assist Video Surveillance-based Vandalism Detection

Tracking #: 1721-2933

Faranak Sobhani
Umberto Straccia
Qianni Zhang

Responsible editor: 
Krzysztof Janowicz

Submission type: 
Ontology Description
Following the exponential deployment of surveillance systems across a wide-spread region of geographic locations, detection and representation of events have become a critical element in automated surveillance systems. We present here an ontology framework for representing complex semantic events to aid video surveillance-based vandalism detection. The proposed ontology builds on DOLCE ontology and relies on the linguistic and cognitive modelling of philosophical knowledge to achieve interoperability between proprietary surveillance systems. The explicit definition of event vocabulary we present here is aimed at aiding forensic analysis to objectively identify and represent complex events. The proposed ontology has then been applied in the context of London Riots, which took place in 2011, and we report also on the experiments conducted to support the classification of complex criminal events.
Full PDF Version: 


Solicited Reviews:
Click to Expand/Collapse
Review #1
By Karl Hammar submitted on 17/Oct/2017
Major Revision
Review Comment:

I have developed this review based on the SWJ reviewer instructions, pasted below for clarity:

"Descriptions of ontologies – short papers describing ontology modelling and creation efforts. The descriptions should be brief and pointed, indicating the design principles, methodologies applied at creation, comparison with other ontologies on the same topic, and pointers to existing applications or use-case experiments. It is strongly encouraged that the described ontologies are free, open, and accessible on the Web. If this is not possible, then the ontologies have to be made available to the reviewers. For commercial ontologies, exceptions can be arranged through the editors. These submissions will be reviewed along the following dimensions: (1) Quality and relevance of the described ontology (convincing evidence must be provided). (2) Illustration, clarity and readability of the describing paper, which shall convey to the reader the key aspects of the described ontology.”

Note for the editors: It may be the case that this article is better reviewed by someone who is an expert on DOLCE. I am not, and therefore have had to read up on DOLCE simultaneously with reviewing the article; consequently, I suggest the editors check those of my comments that pertain to DOLCE issues also with the other reviewers.

The domain and the intended usage of ontology technologies are interesting and I would like to read work in this direction. Unfortunately, however, this article and the proposed ontology has several issues that would need to be resolved for me to consider it acceptable for publishing with SWJ. It is my opinion that a major revision is required (the listed issues could feasibly be fixed within the six week grace period that a major revision decision allows). The most problematic issues are those that relate to the insufficient evaluation of the proposed ontology (below denoted issue 17) and to the lacking narrative (issue 6). Below I have grouped the issues by whether they relate to the ontology or the article.

# Issues about the ontology itself:

Issue 1: After downloading and opening the linked ontology in Protégé, I get the following reported error:

"Level: ERROR
Time: 1508151918747
Message: Illegal redeclarations of entities: reuse of entity in punning not allowed [Declaration(DataProperty()), Declaration(ObjectProperty())]”

Issue 2: The downloaded ontology uses a perdurant subclass naming that is inconsistent with the article: the class that is in the article (and in DOLCE) called Process is in the downloaded ontology named ”Activities”.

Issue 3: The ontology uses many very ambiguous property names, including ”isA”, ”has”, ”isFrom”, ”equals”, ”isIn”, etc.

Issue 4: None of the properties in the ontology have metadata documentation, and several of the classes also lack such documentation, which would be helpful in clarifying intended usage (particularly for properties with ambiguous names, as described above).

Issue 5: The ontology is not published at a dereferenceable namespace. Rather, the article links to a ZIP file with multiple alternative ontology representations within it (and some other article-related data). See and for better ways of going about ontology publishing.

# Issues about the article, its clarity, illustrations, etc:

Issue 6: The overall narrative is lacking. The article begins by discussing the proposed ontology as an implementation of the DOLCE foundational ontology, placing great emphasis on the (honestly somewhat confusing, see comment below) types of events supported by that ontology. It then proceeds to describe how different kinds of criminal behavior are modelled, but this portion does not really make much use of the finer distinctions in DOLCE; instead, the term ”Perdurant” could in several of the examples just as easily be replaced with the much simpler common-sense term ”Event”, and the examples would still work fine. Finally, the article attempts to validate the ontology but as far as I can tell (see details below) this validation is not of the ontology as such, but of a set of developed GCIs.

Issue 7: The literature review (section 2) lists prior work in the field but without really discussing the contributions of that prior work and how it contributes (or not) to the problem domain or how it relates to the work in this article. This section could be extended, to provide a better framing for the developments in this work.

Issue 8: The authors claim that their ontology is an "ontology framework” (section 3, first paragraph). What do they mean by ”framework” in this context?

Issue 9: Why is DOLCE selected as foundational ontology? The authors state (section 3.1) that this is a consequence of DOLCE offering ”a cognitive bias with the ontological categories underlying natural language and human common sense”. I’m not sure that a cognitive bias is necessarily a good thing when modelling ontologies, and even if it were, this sentence does not provide a compelling argument on its own without comparing and contrasting against the other listed ontologies in that same section.

Issue 10: The notions of ”Telic”, ”Stage”, and ”Cumulative” which are used extensively in 3.2, are not introduced or explained anywhere in the article.

Issue 11: Figure 2 is unreadable.

Issue 12: Not being an expert on DOLCE and its endurants/perdurants I may misunderstand this, but to me, it makes very little sense to speak of activities like ”Blaming” or ”Harassing” as states, rather than events (page 5, first column, first paragraph). Similarly, it’s unclear what the authors mean by ”MetaLevelEvent” (page 4, second column, last paragraph).

Issue 13: The first paragraph of Section 3.2.2 content-wise seems to repeat section 3.1. The last sentence of that same paragraph is rather indecipherable.

Issue 14: The DL snippet on page 7 is malformed.

Issue 15: Figure 11 (describing file formats for CCTV footage the authors had access to) and Table 3 (statistics printout from Protégé?) add nothing of value and ought to be removed.

Issue 16: The first column on page 11 seems to describe, in great formal detail, the notions of Precision and Recall. I would assume that these notions be public knowledge that a reader of this journal do not to have need explained.

Issue 17: Section 5 claims to validate the ontology. As far as I understand, this validation only covers whether a) the GCI:s manually created by the authors cover the crime events for which they were explicitly defined (it would be strange if they did not) and b) whether the DL Learner tooling can deduce those same GCIs. This is, I think, insufficient as validation/evaluation of the qualities of an ontology. I would like to see more, for instance, evaluation of the ontology by its intended users (who are those users?), evaluation/validation of the usability aspects (DOLCE and its instantiations are not necessarily the easiest ontologies for non-ontologists to understand, so in case such users are in the target group this might be an issue), coverage versus the ontology requirements, or evaluation of DOLCE as foundational ontology with other possible foundational ontologies and their features, etc.

Issue 18: There are a multitude of small languages issues throughout the text, e.g. mixing up of definite and indefinite articles, overly long and convoluted sentences, etc.

Issue 19: The styling of headers cause clarity issues, e.g., the Section 4.2 header is neither larger in font does it nor style-wise stand out from the surrounding text properly, whereas subsequent labels in that same section appear larger/bolder and look like headers.

Review #2
By Eva Blomqvist submitted on 20/Nov/2017
Review Comment:

The paper suggests an event ontology, based on DOLCE, for supporting video surveillance tasks. The topic is interesting and certainly the need for better automated surveillance is clear. Unfortunately, the paper falls short on a number of other points, some of which are actually requirements of papers submitted in the ontology description paper category. First and foremost, the ontology itself does not seem to be available online. Some experimental material is linked from the paper, but that package does not contain anything that seems to be the actual ontology, and the namespaces referenced in the files that are provided do not resolve. Since this is a requirement for the paper even being considered, this is obviously already a reason for rejection.

However, looking a bit more in detail at the paper, there are also a number of other things that I would suggest the authors to improve, in case they want to publish this as an ontology description in the future.

The first review criteria for ontology description papers is "quality and relevance of the described ontology". While it is hard to say something about the quality of the ontology, without having access to the actual ontology, the relevance could actually also be questioned. The paper does not contain a proper motivation why this ontology is needed, and why existing ontologies are not already able to solve the problems listed here. Section 2 is called "literature review", but I would much rather see a "related work"-section, that really compares the ontologies, rather than simply a list of other literature, without comment or comparison. The only hint in this direction is the first paragraph of the next section, where the authors state that "no standard formalisation of forensic events has been proposed". Does this mean that their proposed ontology would be a candidate for such a standard? Or just that they believe that since no standard exists, it is relevant to propose yet another event ontology? In order for the paper to be useful to the reader, the authors would have to make it very clear why another event ontology is needed, why their proposal fills that gap (compared to other alternatives), and clearly establish the quality of the ontology (i.e., including publishing it online for public scrutiny and reuse).

In the call for ontology descriptions it is also stated that the "descriptions should be brief and pointed, indicating the design principles, methodologies applied at creation". Other kinds of short papers of this journal are recommended to be around 10 pages, so this paper is already a bit long to be called a brief and pointed description, considering its 14 pages. Further, although it is described that the ontology is an extension of DOLCE, there is no description of the development methodology used, and no list of requirements of the ontology (neither in the paper nor supplied as additional material). This makes it hard both to determine the exact scope and intent of the proposed ontology, as well as contributes to the difficulty assessing its quality as already mentioned. From the paper itself it is indeed hard to determine what the scope of the ontology really is - is it to create a comprehensive taxonomy of activities? In that case the scope seems to be huge, in fact is that even possible to achieve? Or is it "only" about forensic perdurant entities? Still, a huge scope in my opinion, and I am not convinced that this can even be done in a general manner, without taking the intended application context into account. Then comes the CGIs, which seem to be even more context and application specific - are these actually intended to be part of the ontology itself? Or are they described in the paper as part of the application case example? This is not clear, however, since one of the experiments is about learning such GCIs I would assume that they are part of the example use case and experimental setup, rather than the ontology.

Continuing to consider the experiments, to me it is not clear what is being evaluated here. In the first sentence of section 5, the authors claim that the experiments are used to validate their ontology - how? I do not see how manually creating, or learning, GCIs is a way to validate the ontology that they are based on. And how can we know if the ontology is any good if there are no requirements or scope for it in the first place?

Overall, and without even mentioning details of the paper sections, I unfortunately find this paper quite confusing. It is not clear what the ontology is for, why it was created, what is its scope, how it was designed, and if/how it was evaluated. Even the ontology itself is not disclosed to the reviewers. All this together makes me suggest that the authors completely rework this paper, rethink its purpose and then start by publishing the ontology, before going ahead and making another attempt at describing it in a paper. Nevertheless, coming back to my initial statements, I am sure that with the right reworking and a much more clear description of the ontology, this could be a very interesting and useful work to share with the community.