Table of Contents
Introduction 3
Background 3
SMART Crane Introduction 4
Research Gap 6
Research Objective 6
Literature Review 7
Background 7
Background Subtraction Technique 7
Chapter 1: Data Cleaning and Structure 8
No table of figures entries found.
Introduction
Background
Based on Central Intelligence Agency(CIA), Singapore has a population density which is 8188.66 per square kilometre and has been ranked top 3 highest in the world. Thus, Singapore government has taken some action to overcome this issue. Singapore government decided to increase additional 700,000 housing units, around 60% increase from today by 2030 to accommodate a population of 6.9 million on 2030 published by Singapore White Paper.
However, Singapore has limited land. High rise building and land reclamation are needed to accommodate the growing population of Singapore. Based on Singapore Land Use Plan published in 2013 by Ministry of National Development, the planned land supply expected to increase from 71000 hectares in 2010 to 76600 on 2030 hectares which is about 8% increment. From the plan, we can see the expected increment supply of land use of 5600 hectares is mainly for Housing, Industry and Commercial. There are 3000 hectares increment by 2030 which has an increase of 3% of total land use from 10000 hectares in 2010 to 13000 hectares in 2030 for housing. As for Industry and Commercial, the planned land supply will be increasing from 9700 hectares in 2010 to 12800 hectares which occupy 17% of total land use in 2030. However, land use for other purposes such as parks, community, institution, recreation facilities, reservoirs, land transport infrastructure, ports and airports will remain constant. From these data, Singapore government has been working hard to provide more housing units to meet the demand of the public.
As a result, Building and Construction Authority (BCA) has to improve the productivity of the construction technology to meet the goals by 2030. This lead to the use of latest automation technology by Singapore government.
Traditional eliminate
SMART Crane Introduction
Kimly Construction has a collaboration research with School of Civil and Environmental Engineering, Nanyang Technological University (NTU) leaded by Professor Robert Tiong to implement a semi-automated precast logistics management & installation system – SMART crane on December 2015 and the implementation was successful on October 2016. This new technology has been tested in one of Kimly’s projects which is Signature at Yishun. SMART Crane technology can enhance the productivity, productivity, efficiency & safety of the precast elements logistics management, hoisting & installation process of housing projects through BIM-GIS/GPS and sensor-based semi-automated navigation system.
First of all, every precast element that are going to be used must be embedded with a Radio Frequency Identification (RFID). It is an ID system that uses small radio frequency devices for identification and tracking purposes. An RFID tagging system includes the tag, a read/write device, and a host system application for data collection, processing, and transmission. Thus, every single precast element that are embedded with the RFID can be easily to be track by contractor.
Figure 1: RFID Tag
Once the precast element has been delivered to site, the workers must scan the elements. This is to allow the team to track the location of the elements and for the documentation of the inventory.
Figure 2: RFID Scanner
Subsequently, the tower crane tracks the location of the specific precast element and the tower crane operator just needs to follow the path created by the system and hoist the items to the designated location. The worker must scan the RFID again when the elements have reached the target location and the installation work has started.
Figure 3: RFID Tracking Panel
The construction progress can be tracked easily and will be sent to BIM team for record.
Research Gap
However, there are some limitations by using SMART Crane compared to traditional tower crane. Traditional tower crane is manually controlled by tower crane operator while SMART crane will calculate the optimum hoisting route for the tower crane operator to follow. Thus, some safety issues may arise as the route calculated by SMART crane does not take into account the existence of manpower.
Research Objective
The objective of this study is to develop an algorithm to detect manpower and signal the manpower to stay outside the danger zones within the SMART Crane hook range. A few series of analysis are to be carried out to get a best result of the detection. This algorithm will help to eliminate the safety issues that may arise from the automation of SMART Crane and enhance the SMART Crane.
Literature review
Background
Object detection and tracking has been a norm nowadays. It has been used in various industries such as farming, military, transportation, video games, security and so on. There were several techniques that have been used for object detection and tracking in the past such as Background Subtraction Method, Real Time Background Subtraction and Shadow Detection Theory, Template Matching, Image Differencing, Shape Based, Optical Flow, Frame Difference, Motion Based, Texture Based, Colour Based. The effectiveness of each tracking method is categorised based on processing speed, memory requirements and accuracy. There are also some conditions that can affect the effectiveness of the detection and tracking such as weather, light intensity, disturbance, overlapped object and moving speed of the object. The selection of object detection and tracking techniques is to choose the one with best performance and fastest response time.
Mechanism
Image processing can be divided into 2 groups which is, human detection and tracking with two classifiers shown in [1]. There are three stages in video analysis: interesting object detection, interesting object detection and activities recognition.
Video will be divided into each frame depends on what is the intervals to be taken
DIAGRAM
Background Subtraction Technique
Background Subtraction Algorithm had been commonly used in [1] for human motion detection from a static camera. The basis in this method is to detect the moving objects from the changes between the current frame and a reference frame to detect the motion region which is called the “background copy” and “background replica”. In the background image, there is no moving object and must be kept updated consistently to accommodate to different geometry settings and luminance conditions. The fundamental concept of this method relies significantly on initialisation and background image updating. Thus, the accuracy of the detection will be highly depending on the both concept stated above.
Algorithm process is shown as below:
• Video frame sequence
• Separation of frame
• Image sequence
• Existing frame image and background frame image separation
• Background subtraction operation
• Moving object detection
• Background updating operation
• Update of background
• Removal of noise
• Analysis of shape
A. Background Image Initialization
There are a few ways to get the initial background image. For example, the first frame of the video can be taken into as the initial background or the average pixel brightness of the first few frames can be taken as the background. Average method has been commonly used for background initialization among these methods. However, there will be one major disturbance by using the average method which is the shadow problems and can greatly affect the accuracy of the result in the later stage. Thus, the shadow problems can be resolved by using the median method.
Expression of median method:
Binit (L,M) = median Fk(l,m) k =1,2 ….n
Where:
Binit = initial background
n = total number of frames selected.
B. Moving Object Mining
Interested objects in a frame will be filtered out through background subtraction method. Images that contains the object will be subtracting out compared to the previous background image which do not have any foreground interested objects. The pixel location of the moving objects will be affected by the selection of area of the image plane. Thus, Threshold technique are used to separate the moving objects that formed by a group of pixels from the background image.
Background image B(l,m) is then subtracted from the current frame Fk(l,m) to get the pixel difference between these two frames. The pixels of the moving object are then being detected if the pixel difference is greater than threshold T, and the pixel difference that is smaller than the threshold T will be categorized into background pixels. Finally, the moving have been found after going through the operation.
Expression of the threshold operation:
Dk(l,m)= 1 if Fk (l,m)-Bk(l,m) ≥ T
0 Otherwise
Where:
1 = moving objects pixels
0 = background pixels
However, the moving objects may not necessary to be a human body. Thus, further filtration is needed to capture the human body. The further technique to be used is based on the shape features to determine the human bodies among those moving objects.
Flow chart of moving human body mining (extracted from …….)
C. Noise Removal
There is a lot of noises contained in the motion region as well. The factors that contribute to the noise are environmental factors, illumination changes and so on. Thus, median filter had been used to remove the noise with the 3 x 3 window for noise filtration.
Some methods in this technique have been used which are frame difference technique, real time background subtraction, shadow detection technique, adaptive background mixture for real time tracking technique.
Image processing
Traditional object tracking
Artificial intelligence
Chapter 1: Data Cleaning and Structure
Component 1: Construction industry. Safety. Smart Crane history. Construction industry
Component 2: Safety. Research Gap – things that not been done in the past, propose extension of the research
Component 3: Increase safety awareness. State the objective and hypothesis of my study, outline main feature/scope, explain significance
http://www.ifs.du.edu/ifs/frm_CountryProfile.aspx?Country=SG
https://www.mti.gov.sg/MTIInsights/Documents/FAQs%20for%20White%20Paper.pdf
https://www.mnd.gov.sg/landuseplan/
Abstract
Introduction
Literature review
Chapter 1: Data Cleaning and Structures
Chapter 2: Feature Extraction
Chapter 3: Object Detection Using YOLO
Chapter 4: Lifting Safety Distance Modelling
Methodology
Understanding of R-CNN
So, to summarize, R-CNN is just the following steps:
1. Generate a set of proposals for bounding boxes.
2. Run the images in the bounding boxes through a pre-trained AlexNet and finally an SVM to see what object the image in the box is.
3. Run the box through a linear regression model to output tighter coordinates for the box once the object has been classified.
R-CNN works really well, but is really quite slow for a few simple reasons:
1. It requires a forward pass of the CNN (AlexNet) for every single region proposal for every single image (that’s around 2000 forward passes per image!).
2. It has to train three different models separately – the CNN to generate image features, the classifier that predicts the class, and the regression model to tighten the bounding boxes. This makes the pipeline extremely hard to train.
Source: CIA World Factbook – Unless otherwise noted, information in this page is accurate as of January 1, 2018