As a continuation to one of my previous posts, here I am going to discuss some specific use cases.
Vehicular Fog Computing
With the growing adoption of dash cameras, we are seeing great potential for innovation by analyzing the video collected from vehicles. As we previously agreed, transmitting and analyzing large amounts of video requires a lot of communications and computing resources. However, vehicular fog computing for real-time analytics can be applied to mitigate this issue. This can also be applied to smarter vehicles that are equipped with onboard sensors and cameras.
Vehicular crowd sourcing is the process of video crowd sourcing in vehicular fog computing. The whole process consists for four operations:
- Discovering fog nodes: a vehicle needs to figure out which vehicular fog nodes are located within its communication range.
- Sending requests: after discovering fog candidates, the client sends a request to the zone head over using LTE.
- Collecting dash camera video: the description of the generated dash camera video, such as length and supported video resolutions.
- Conducting vehicular application service: after collection of the dash camera video, specific vehicular application service would be conducted in the fog nodes.
Here are some examples of such applications:
• Driving assistance. This can include driving situational awareness, such as cooperative lane changing, and see-through for passing.
• Local 3D map generation. Autonomous rely extensively on high-definition 3D maps to navigate. Crowdsourced dash camera video can be utilized in construction of such maps, as well as their updates.
• Crime Scene Reconstruction. Reconstruction of a crime on the basis of evidence is very important for law enforcement. However, criminals like to destroy or obstruct local CCTV security cameras. In such case, a dash camera approach can provide necessary resources for crime scene reconstructions.
Some other applications include parking navigation, road construction detection, infrastructure improvement recommendation and many other.
To manage the increased volume of data from connected and autonomous car appliances that often generate more upstream data than downstream, reverse Content Distribution Network (rCDN) concept is applied. The key features of rCDN are:
- Dynamic video splicing. Each rCDN node receives multiple video streams from multiple cameras or downstream rCDN nodes and splices them together based on relative likeness of scenes.
- Dynamic and adaptive transcoding to match IoT service needs and rCDN node capacity. Service-aware and server-aware video management mechanisms through dynamic and adaptive video transcoding pre-caching and post-caching. This requires the video analytics to examine the video chunk size and the associated meta data.
- Video content upstream distribution. Each rCDN node can obtain information from upstream rCDN nodes on their locations, remaining caching space, and type of video chunks cached.
- Dynamic retention policies to match IoT service needs. rCDN nodes periodically examine the lifetime of each cached video chunk, its size and associated metadata and determines the retention priority and policy, which may include data migration.
- Multiple caching realms creation. Creating multiple caching regions within each rCDN node allows for faster cache hits, parallel and multiple kinds of access by different service to the same rCDN node.
Urban surveillance is an essential part of situational awareness for better urban management and planning, deals with heterogeneous data from a layered sensors environment. Efficient extracting, analyzing and understanding the large-scale data set from heterogeneous smart devices in a real-time manner are essential for critical and emergency situations. However, there is still a huge performance gap between the amount of data generated and the lack of adequate resources to process it.
For urban surveillance tasks requiring complex data fusion, cloud computing has been widely recognized. However, cloud computing does not work for all kinds of applications. The extra round-trip delays and possible network congestions are not tolerable. In this case, fog computing can be a promising solution.
Smarter cities utilizing IoT technologies are required to provide a sustainable environment to accommodate the needs of the increasing urban population. A s mart city goal is to improve quality of life for its citizens. This objective can be achieved by deploying sensors, such as cameras, across the city to analyze public area crowds in order to improve city management.
Because cities are big with large public areas, it means that there is a challenge to manage the quantity and variety of potential sensor data, including video. The tree major issues for such a case are:
- Optimizing network bandwidth
- Real-time responsiveness
- Personal data privacy preservation
Just like in surveillance case, cloud computing might not be the best model to manage and process all the data. This is where implementation of fog computing can solve multiple issues. Highly congested areas of large cities present many challenges for city planners. Monitoring crowd movement can help understand crowd patterns to avoid congested areas and ensure that people move safely, securely, and in predictable manner. This has the potential to improve quality of experience and safety.
Crowd monitoring uses image processing. Data captured by the video cameras is highly sensitive, and therefore outsourcing to cloud computing for processing is not suitable. However, the image itself doesn’t matter, but the aspects of metadata like crowd density, activity, or the number of people present matter. Because of this, video processing in fog computing improve responsiveness while ensuring data privacy.