QuiltView: a Crowd-Sourced Video Response System

Motivation:

Portable glass like devices such as the Google Glass can prove to be highly useful and efficient as they facilitate the implementation of the crowd-sourced systems. The unique and effortless one-touch video capture by the Google glass can be used by these type of systems to receive queries and respond to them accordingly. Video responses carry highly rich information and give the user a sense of the actual context that allows him to judge the situation as if he is actually seeing it live.

Main Points:

• This paper proposes a cloud-based service, QuiltView, that processes short queries with deep semantics and returns brief video segments rather than to give detailed verbal responses.
• QuiltView takes advantage of the specific strengths of the Glass devices, such as micro-interaction, that causes low user distraction, and low cognitive load for this style of interaction.
• QuiltView has a global catalog of users and queries implemented with a SQL database. It includes the details of all the queries and the responses that have been made. Also, the catalog holds the details of the users preferences regarding their willingness to respond to the queries.
• The uploading and viewing of videos are done using standard YouTube mechanisms that are wrapped inside QuiltView query and response software. Each video clip is uploaded into YouTube, and its link is displayed in the list of  responses to the query. Meta-data about this video clip, matched to the original query, is entered into the global QuiltView catalog.
• The QuiltView architecture incorporates result caching, geolocation and query similarity detection to shield users from being overwhelmed by a flood of queries.
• It uses a query matching algorithm that detects the queries that are semantically same. This ensures that the cached queries are used optimally because the chances of two users passing the same query with similar literals is very low.
• Since the user preferences about receiving queries may vary considerably, depending on the individual, time, location, query topic, and reward offered, the QuiltView before sending the queries to users creates a set of eligible users based on their preferences.
• The paper provides a couple of use cases One of them is to handle traffic emergency where the police sends a query asking the scene of the blockage. The drivers and passengers respond with short videos and the emergency response center can take decisions accordingly. Other use cases are scavenger-hunt game and free food finder where this kind of system can prove useful.

Trade-offs

• This system has high power and high bandwidth requirements. The authors have tried their best to reduce unnecessary video uploads by caching and query matching. Anyhow, the users must be aware of the system requirements and the costs that may incur by using this system.
• Though having restrictions on the query length provides good user experience, it would be quite difficult for the user who must add details to queries to receive correct video responses. So, I feel there must be flexibility on the length of the query. One way I can think to fix is, like other user preferences there must be another preference added that says if long queries allowed. 
• This system puts a restriction on the size of map zoomed-in region. This is not the best system for queries that are not location specific and can be responded by anyone and from anywhere in the world. There must be a query analyzer that can differentiate between generic queries and location specific queries. The restriction must be imposed on the number of users to which the query is sent in case of generic queries.
• The response videos are uploaded to YouTube. Thus if the videos include recordings of a third person's activity then this infringes the privacy of others.The authors doesn't provide convincing arguments regarding this.
• The responses received by the user should have a correctness parameter added to the links. This can be done by pre-processing of videos in the cloud and labeling the videos as which one would be a best response for the initiator. This saves unnecessary bandwidth and power usage because if the user is satisfied by one video, he would not stream other videos.
• Since this system uses YouTube, it has to comply to all the policy guidelines of YouTube about the contents of the video and the broadcasting the video contents to various locations.