Video upload from public transport vehicles using multihomed systems

We consider a surveillance system for public transport vehicles, which is based on the collection of on-board videos, and the upload via mobile network to a central security system of video segments corresponding to those cameras and time intervals involved in an accident. We assume that vehicles are connected to several wireless interfaces, provided by different Mobile Network Operators (MNOs), each charging a different cost. Both the cost and the upload rate for each network interface change over time, according to the network load and the position of the vehicle. When a video must be uploaded to the central security, the system has to complete the upload within a deadline, deciding i) which interface(s) to use, ii) when to upload from that interface(s) and iii) at which rate to upload. The goal is to minimize the total cost of the upload, which we assume to be proportional to the data volume being transmitted and to the cost of using a given interface. We formalize the optimization problem and discuss greedy heuristics to solve it. Then, we discuss scientific and technical challenges to solve the system.


I. INTRODUCTION
We study a video surveillance system for public transport vehicles, which is based on the collection of on-board videos and their wireless transmission to a central security system.Our interest is motivated and inspired by the real needs of public transport operators.On public transport vehicles, already now, several video cameras are installed, each producing a video stream with rate from 1 Mb/s to 10 Mb/s.Continuous real-time video streaming from vehicles to the central security system is considered too expensive in data volume and in cost, and largely useless, because nothing relevant happens on the vehicles most of the time.Videos are thus stored on board, and when an alarm is triggered (e.g., when a customer or a driver reports a problem, or after a complaint is filed), the Security Operator (SO) on duty in the central security control station needs to access the portion of the on-board videos which refers to the period of time of the accident.In traditional systems, videos are uploaded to the central security system when the vehicle enters the depot, where cheap and high-speed wireless connectivity is available.This forces the SO to wait a long time before being able to investigate the accident.
Here, we consider a novel solution, which provides the SO with near-real-time access to videos corresponding to those cameras and time intervals involved in the accident.We assume that the vehicle is connected to the network by means of different wireless interfaces, through different Mobile Network This work was funded by the MONROE project (grant agreement no.644399) in the H2020-ICT-11-2014.
Operators (MNOs), each charging a different cost, from cheap WiFi, to 3G/4G interfaces, or satellite links.Both the cost and the upload rate for each network interface change over time, according to the network load and the position of the vehicle.We assume that, thanks to the repetitiveness of the public vehicles routes, the system can create a performance map to collect information about the expected network connectivity performance along the route.(The creation of such map is outside the scope of this paper.) Once the SO requests a video, the system has to complete the upload from the vehicle storage system within a given deadline.The system has to decide i) which interface(s) to use, ii) when to upload from that interface(s), and iii) at which rate to upload.The goal is to minimize the total cost of the upload, which we assume to be proportional to the data volume being transmitted and to the cost of using a given interface.For instance, assume that a video must be uploaded with a deadline of 5 minutes, and that the cost of using a given operator (slow and expensive) 3G interface is higher than the cost of using a (fast and cheap) WiFi interface of a second operator.However, the bus will enter the coverage area of the latter only in 3 minutes.In this context, is it better to wait entering under WiFi coverage, or to start uploading the video now?
The video upload problem can be seen as an optimization problem for which it is possible to obtain different formulations, depending on the assumptions.We discuss possible formulations and heuristics to solve it.Next, we faced the system design challenges that must be solved when engineering the entire system, like how to estimate and predict capacity, impact of transport protocol, impact of uncertainty, etc.

II. MODEL FORMULATION
We model the scheduling problem using a directed graph G = (N, E), where N = {i} is the set of nodes and E = {(i, j)} is the set of edges.Referring to Fig. 1, the leftmost node represents the video source, i.e., the vehicle.The second group of nodes represents the video files to be uploaded.Each video k (2 videos in the example) is of volume V k , and can be uploaded through different interfaces, at different time slots, represented by the third group of nodes.Each node in this group represents a given interface and time slot.For ease of visualization, nodes referring to the same interface (2 interfaces in the example) are grouped by a box.The number of available time slots (5 in the example) represents the deadline to meet (recall that we consider slotted time).The rightmost node represents the sink, i.e., the server receiving the videos.
All edges in E have a label containing two values: a cost and a capacity.The label of edge (i, j) is denoted (c i, j ,r i, j ).