Uav Surveillance for Urban Crime Prevention

According to the statics, the worldwide Unmanned Aerial Vehicle (UAV) or drone market is expected to reach 41.3 billion US dollars by the year 2026 [1], this indicates the growth and usage of drones in various fields, these drones changed our lives significantly over the past few years. But still, we are not using these drones for surveillance. The crime rates that happen in the back alleys are still higher than the other places in the urban cities because most of these parts are not under CCTV surveillance also it can patrol the areas which are expected to have more crime rates or private properties to prevent trespassers. It can also increase women’s safety at night who work late. GCS is open-source software that can be downloaded from the internet and used for mission management and real-time flight operation. There are eight open-source projects for UAVs, we discussed five autopilots such as processors, gyroscopes, accelerometers, magnetometers and barometers. Introduced flight controllers with various functions to monitor and control UAVs. Their additional functions can be added by extending the open-source code. Mission planners and Q-ground control are known as open-source software packages that deliver UAV flight automation.

It enables UAVs to fly along with advanced predetermined Waypoints and organize realtime flight data. It has the function to return whether the battery is low. The functionality of GCS is convenient due to its wide range. There are some functions such as surveillance and object detection such as people, knives, and guns are neither available. There is too much attention on flight functions and the difficulty of making UAVs for Non-traditional tasks convenient for normal users. The making of customized GCS is a project requirement as mentioned. In realtime, the multiple UAVs are monitored by the development of an efficient and comprehensive guidance system, and the main layout is also discussed. Customized GCS are developed using an open-source library. Here in this research, unmanned helicopters with a two-layer framework that which data are transferred in the background are developed in the real- time software system. The customized GCS surveillance system is inspired by light radar. The real-time tracking of commercial flights by providing geographical locations and flight data. To receive the flight data and command of the UAV use the flight numbers, speeds and MAV link protocol.

The proposed surveillance system is (1) the monitoring system that provides flight data. (2) Autonomous flight controlling, (3) real-time monitoring, and (4) area monitoring, (5) detection of weapons or trespassers, (6) patrolling at regular intervals. In this system, the #DR telemetry (TX/RX) is employed to attach the monitoring system, and also the Pixhawk supports PX4 firmware and ardupilotmega firmware. The UAV ID in MAVLink is used for the specific system IDs assigned for each autopilot. MAV Link messages and data carried by the telemetry, actual system ID of the XML record contain the data of the flight data and UAV GPS information. At the same time, the UAV interaction in functions like take-offs, landings, and auxiliary servo moments are worked by the commands of the monitoring system. The XML file is also used to send commands to the UAV. The MAV Link C header library is used to generate the XML files. JavaScript package manager provides NodeJS MAV Link to use decoding or encoding. TX and RX are two telemetric and send telemetry messages by (Serial COM6) and receive messages by (Series COM5).

Proposed system

The surveillance system we propose here can be separated into two sections (1) Algorithm for object detection, and (2) flight planning of the UAV. The object detection algorithm implemented in the system constantly monitors the area under the surveillance for the trespassers in case it used on private property and weapons such as guns or knives in the ordinary street surveillance it also sent an emergency signal in case it been attacked or damaged to the police or the person in charge. In path planning, the UAV is assigned the path it needs to follow and it also gives it the instruction to take on or take off. Using this path planning the drone follows the path and returns to its station once the patrol is completed.

In this system, we use 3DR telemetry (TX/RX) to attach the Pixhawk autopilot UAV and also the monitoring system. The Pixhawk supports PX4 firmware and also the Ardupilotmega in our system [2-10]. We are using Ardupilotmega in this system. Each autopilot is assigned a specific system ID and they are assumed because of the UAV ID in MAV Link. The messages and data are carried by telemetry in MAV Link, this data contains the UAV GPS information and flight data of the actual system ID of the XML record. In the meantime, the monitoring system sends a command like take-offs, landings, and auxiliary servo movements to the UAV to have interactive functions. The commands are also sent to the UAV through the command XML file. XML files are created by the MAV Link C header library, which can be decoded or encoded using Node.js MAV Link of the JavaScript package manager. Here we are using two telemetries-TTX and RX. One telemetry (serial COM6) is employed to send while the opposite (serial COM5) continues to receive telemetry messages.

Tables 1 & 2 refer to the MAV Link messages and commands that are used to execute the monitoring system. Table 3 & 4 shows us the report status messages (failed or successfully executed) that appear after the commands are sent (Figure 1).

Table 1: MAV Link Message.

Table 2: MAV Link Command.

Table 3: Acknowledgment message.

Table 4: Flight Mode.

Figure 1: Assigning two waypoints with Node.JS MAV Link.

The weapon detection algorithm was designed using the application BagOffeatures and the category Classifier system. The analysis of the weapons is made into a combined input and fed into the system This phase receives the raw images from the drone, extracts the features, trains, and tests the system with the data. fist need to image pre-processing. The image set used in this research contains three sets of images: Airplanes, Guns, and Knives The different sets contain different amounts of images the image sets contain images whose object of interest seems to have a conflict with the background object The obtained results then, can be compared favorably with those of electromagnetic wave images. the electromagnetic waves when passed through a closed space, it acts as a MOM (Method of Moment), whose functions are unable to analyze metallic objects. Feature extraction. The next phase of image processing which is the taking out of peculiar features of the various sets is done using the BagOfFeatures function. Matrix laboratory (MATLAB) 2020 functions are used to flawlessly extract the features of the different groups. it identifies some identical features of the group which it considers strong features [11-17].

The strongest features consist of those characteristics that distinguish one category from the other. r it selects 80% of the strongest features, and this increases its sensitiveness when objects in such categories are tested with the system. The 80% strongest features extracted also help to overcome the challenges that may arise from blurry images from drone cameras. BagOfFeatures goes further to harmonize the number of these strongest features amongst the three groups. For the proposed system, the least number of strongest features have been gathered from the gun’s category, with 12,446 strongest features. Hence the 12,446 strongest features of all three image sets are chosen and saved in the BagOfFeatures functions. K-means clustering is then implemented to create a vocabulary consisting of 100 visual words. This is a method of vector quantization. For the proposed system, a total number of 37,338 features were applied in this stage, and 100 clusters were created. The iteration in the stage was covered after the 14 iterations, at a 0.08 s/iteration rate. For each pixel of any image, it computes the Euclidean distance (d), between the center of the image and each pixel of an image using the function in the equation below. d = p(x, y) − Ck (1) where, k is the number of clusters, x and y are the dimensions of the resolution of the image, and Ck represents clusters center the visual word occurrences (VWO) created for the three different groups. The guns category has a fair amount of VWO, due to its range of unique features.

The image category classifier function is used to classify images. In this stage, the output of the BagOfFeatures becomes an input. The function then trains a Support Vector Machine classifier (SVM) using the BagOfFeatures input. The parallel computing capabilities of this function make it possible for the system to train with as many features as are fed into the system.

where x is an observation (corresponding to a row of X). b is the bias term (corresponding to MDL. Bias). the trained system is evaluated using three sets of test categories. The test sets are a combination of images of different objects. The ability of the system to detect and correctly identify the number of each object in each of the sets displays the sensitivity of the system to any objects that may be introduced into the system. the results of single image test sets are easy to read. Hence we applied the same matrix order in introducing the test sets. From the final result, we see that 88 images were tested for each of the test sets and the confusion matrix shows a 95% accuracy for knives, 80% accuracy for guns, and 73% accuracy for airplanes (Tables 5 & 6). The resulting confusion matrix shows an average accuracy of 94.67% for the overall system. Hence, we can consider the system to be reliable to the nearest 0.01 error rate.

Table 5: The results of the first evaluation and test.

Table 6: The results of the evaluation and test validation.

The UAV is capable of monitoring and surveillance the urban streets without causing any major changes to the lifestyle of the people. Though this method is widely used during the crisis of COVID-19 to monitor the implementation of lockdown all are operated using operating personnel. But this method reduces the operating personnel as well as increases the safety of many citizens. An average commercial drone can fly for about 15-20 minutes when fully charged we use this time for patrolling and recharging once it reaches the station it can recharge itself and continues its patrol at regular interval. At the present study, our surveillance system can monitor around 300m of radius without any major issues

1. https://www.statista.com/statistics/1234521/worldwide-drone-market/
2. Lim H, Park J, Lee D, Kim (2012) Build your quadrotor: Open-source projects on unmanned aerial vehicles. IEEE Robotics & Automation Magazine 19(3): 33-45.
3. Khalid A, McFall K (2014) Aerial robotic autonomous patrol and surveillance system. 14th AIAA Aviation Technology, Integration, and Operations Conference 1-10.
4. Perez D, Maza I, Caballero F, Scarlatti D, Casado E, et al. (2013) A ground control station for a multi-uav surveillance system. J Intelligent Robot System 69: 119-130.
5. Dong M, Chen BM, Cai G, Peng K (2007) Development of a real-time onboard and ground station software system for a uav helicopter. Journal of Aerospace Computing, Information, and Communication 4: 933-955.
6. http://refhub.elsevier.com/S0045-7906(19)30393-3/sbref0006
7. https://www.npmjs.com/package/mavlink
8. http://mavlink.org/messages/common#MAV_CMD_ DO_SET_SERVO
9. http://mavlink.org/messages/common
10. http://qgroundcontrol.org/mavlink/waypoint_protocol
11. Giyenko A, Palvanov A, Cho Y (2018) Application of convolutional neural networks for visibility estimation of CCTV images. In: International Conference on Information Networking, Chiang Mai, Thailand, pp. 875-879.
12. Gupta BB, Quamara M (2018) An overview of internet of things (IoT): Architectural aspects, challenges, and protocols. Concurr Comput Pract Exp 32(21): 1-2.
13. Nur Ibrahim AW, Ching PW, Gerald Seet GL, Michael Lau WS, Czajewski W (2010) Moving objects detection and tracking framework for uav-based surveillance. 2010 Fourth Pacific-Rim Symposium on Image and Video Technology, Singapore.
14. Ali S, Shah M (2006) Cocoa-tracking in aerial imagery. Airborne Intelligence Surveillance Reconnaissance (ISR) Systems and Applications III 6209: 62090D-62096D.
15. Harris C, Stephens M (1988) A combined corner and edge detector. Alvey Vision Conference 15: 147-151.
16. Fischler MA, Bolles RC (1981) Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography. Communications of the ACM 24(6): 381-395.
17. Bradski G (2000) The openCV library. Doctor Dobbs Journal 25: 120-126.