Cleanliness Assessment Of Urban Streets With Mec And Dl

Abstract

During the process of smart city construction, city managers always spend a lot of energy and money for cleaning street garbage due to the vast appearances of street garbage. Consequently, visual street cleanliness assessment is particularly important. However, the existing assessment approaches have some clear disadvantages, such as the collection of street garbage information is not automated. To address these disadvantages, this paper proposes a novel cleanliness assessment approach of urban street using mobile edge computing and deep learning. First, the high- resolution cameras installed on current poles is used to collect the street images. Mobile edge servers are used to store and extract street image information temporarily. Second, these processed street data is transmitted to the cloud data center for analysis through city networks. Faster Region-Convolutional Neural Network (Faster R-CNN) is used to identify the street garbage to categorize what time of garbage and count the number of garbage. Finally, the results are incorporated into the street cleanliness calculation framework to visualize the street cleanliness levels, which is convenient for city managers to arrange clean-up peoples effectively.

Introduction

A smart city is an urban area that uses technologies such as the Internet of Things (IoT) Cloud computing [2] and other information technologies to manage and assess the resources and environment of a city in an efficient way.

The smart city concept integrates infor- mation and communication technology, and various physical devices connected to the network to optimize the efficiency of city operations and services [4]. Street cleanliness [3] represents the spiritual outlook and humanistic atmosphere of a city. Keeping the streets clean is good for the development of modern cities. Currently, many major cities regard urban street cleanliness as one of the primary tasks of urban civilization [7]. The European city cleaning network summit also points out that cleaning streets timely is an effective way to improve city cleanliness [6]. At present, the large number of streets make the amount of garbage[8] on streets uncontrollable. Meanwhile, the process of garbage detection on streets is not automated and always requires human intervention at almost every level [4]. Citizens check the location of garbage manually and submit reports to city administrators, then city administrators arrange nearby city personnel to sweep garbage. Some cities even set up cameras at the crossroads of the streets to see if there is any garbage in the area. However, these manual solutions cannot grasp garbage cleanliness of all the streets of the city in time. For this reason, researchers [1], [2] around the world are studying automated approaches, using a cleaning[4] vehicle with cameras to capture the streets regularly and collect street information, such as street pictures, geographical location, date and time. Besides, existing object detection algorithms are used to detect images in the remote cloud platform. Finally, the detection results are sent to the city managers for decision making.

In summary, the main contributions of this paper are described as follows:

1. We describe a novel edge computing framework. An edge layer between cloud servers and mobile terminals. We configure edge servers (micro-data centers) to handle a part of services from mobile devices at the edge layer. It can also store data resources temporarily and transmit data resources in time.
2. Faster R-CNN is used to identify street garbage categories and count the number of garbage. A multi-layer assessment model across different layers is used. The whole city is divided into 5 layers: city, area, block, street, point.
3. A public garbage data set is collected by our- selves, which can be used as a benchmark for evaluating street garbage detection and street cleaning The application validates the feasibility and usability of the proposed approach. The results are useful for improving and optimizing city street cleanliness.

Preliminaries

Mobile Edge Computing

With the rapid construction of smart cities, the Internet generates a large amount of data. Traditional cloud computing requires that data must be transmitted to the cloud center for centralized processing. Remote cloud is a smart brain for processing big data [31]. Since the cloud center is usually far away from end users, it is largely unable to provide low latency. In order to solve this problem, mobile edge computing has been proposed to deploy computing resources to devices close to the terminal. The European Telecommunications Standards Institute (ETSI) [9] definesmobile edge computing (MEC) as a distributed mobile cloud computing (MCC) system. The computing resources are close to mobile devices, and functions such as computing, storage, and processing are added to the wireless network side. In fact, mobile edge computing is based on cloud computing. It only calculates a small part of service. It is especially important for big data analysis. For example, when a user uploads a video or makes a comment, he/she can send it to a remote server through an edge virtual server. The edge virtual server can extract the video content and estimate the possibility that other people want to watch the video. If the probability is high, the edge server will cache this video locally so that anyone interested in this video can get the video directly from its cache instead of receiving it from a remote server, which saves transmission resources and reduces latency. In this paper, we use mobile edge computing to process street images in advance and filter out pictures that meet our needs, which has a good effect on recognition efficiency.

Multi-Level Assesment Model

To measure the cleanliness of the urban streets, our street cleanliness assessment approach provides a multi-level assessment model across different layers. This model can be divided into five layers. Layer 1 is the first layer, it is defined as the city area and sets the scope of assessment. Layer 1 covers all the streets in the city. Layer 2 is the second layer where a city is divided into multiple areas and each region is an administrative area. Layer 3 is the third layer where each area is divided into multiple blocks accord- ing to the sub-administrative area. Each block is uniquely identified by a combination on administrative area and block name. Layer 4 is the fourth Layer where each block has several streets. Layer 5 is the bottom layer where each street has several data collection points.

Deep Network

Deep learning originates in artificial neural networks. By establishing multiple hidden layers and training large amounts of data, useful features can be learned to achieve the expected classification effect.A deep learning object detection algorithm called Faster R-CNN based on region proposal. The algorithm has two main modules: the Region Proposal Network (RPN) proposal box extraction module and the Fast R-CNN detector module is a proposal detector based on RPN extraction and it identifies the object of the proposal box. RPN shares the same convolutional layers by using a convolutional neural network based on object detection and a convolutional neural network that generates a suggestion window.

The feature map extracted by the shared convolutional layer generates a suggestion window through RPN net- work, and gives region suggestions and region scores; The feature map of the first step is input to the pooling layer in Fast R-CNN to extract area features. Combined with region suggestions and region scores, classification probabilities and bounding box regression are trained, the classification scores of the region are output, and the results are finally tested. Faster R-CNN is considered as one of the most precise image detection approaches. It has high detection accuracy and speed. Consequently, the street garbage detection approach in this paper adopts Faster R-CNN (Regional- Convolutional Neural Network) as the underlying model to detect the type and quantity of street garbage.

Approach Overview

The approach is mainly composed of three parts, as described in the following:

1. The first step is data collection and scheduling feed- back in the local management. The city administrators control the mobile station to collect the street garbage image data and respond to the level of street cleanliness presented by cloud center in real time. Then municipal cleaning personnel is arranged nearby.
2. The second step is called data preprocessing. During this step, we use the edge server to store the image data captured by the mobile station temporarily and carry out road judgment of the images from the mobile station in advance. Then, the edge server filters out the images containing road areas. We use linear normalization to get the same size images and these images are sent to the cloud center for garbage detection.
3. The third step is the model establishment and cleanliness calculation. During this step, the cloud server provides an object detection algorithm. Then a model is trained by selecting appropriate parameters and iterations to detect garbage on the street. In the garbage detection stage, we design a counting function to count the quantities of garbage detected. Finally, based on the results of the above detection, street cleanliness level is calculated with respect to different levels.

Literature Survey

In [1] The framework provides an idea for realizing sustainable development of a smart city. The conceptual model could also be used to synchronize and optimize city’s investments. We have studied about practical usage of smart cities. The proposed vision is achieved by providing a common access mechanism to the heterogeneous data sources offered by the city, which reduces the complexity of accessing the city’s data whilst bringing citizens closely to a prosumer (double consumer and producer) role and allowing to integrate legacy data into the cities’ data ecosystem. This system is proposed to classify waste in an automatic way as an application of computer vision in Colombian high schools. Computer vision system classify waste automatically in three modules 1. image acquisition, 2. image processing 3. robotic modules.

The purpose of this research paper is to propose the Anti-Litterbugs Campaign as a more viable alternative to improve and maintain urban cleanliness. The proposed a system examines initiatives by the UK national government to facilitate urban technological innovation through a range of strategies, particularly the TSB Future Cities Demonstrator Competition. The cleanliness status of streets is collected using mobile stations connected via city network analysed in cloud and presented to the city administrator. This paper proposes combining Smart City and Lifecycle concepts to improve vertical service provisioning and horizontal integration between different sectors, across different phases while creating a suitable platform for information and knowledge sharing within the same event and with other similar events. Disposal and beneficial-use options for street sweeping residuals collected as part of routine roadway maintenance activities in Florida, USA, were assessed by characterizing approximately 200 samples collected from 20 municipalities. This paper develops a conceptual framework to examine and analyse two leading cases from the US and Asia. Through the lens of this new framework the paper identifies heterogeneous and heterogeneous characteristics in the process of planning and developing a smart city.

Conclusion

The development of novel technologies has driven a number of cities into the way to smart cities. Street cleanliness is one of the concerns for smart cities. Consequently, this paper proposes a novel urban street cleanliness assessment approach using mobile edge computing and deep learning.