Cadiz, Marc James Andrew E.; Catchuela, Jewel James M.; Condes, Marielle S.; Franco, Reynold M.; Pascual, Sophia Anne C.
Dalis-AI: An AICCTV integrated littering detection system using yolov5 algorithm for Barangay 439, Sampaloc, Manila
ABSTRACT: STATEMENT OF THE PROBLEM: One of the causes of the widespread garbage problem is improper waste disposal which starts in small-scale communities called barangays. It can be observed that most people often throw their garbage in the streets. Some residents are unaware of the consequences of their actions and continue not to follow the rules regarding proper waste disposal. Aside from that, because of the lack of discipline of the residents, littering results in adverse effects on the community, such as clogging of drainages, unpleasant scenery, and harmful effects on the health of the residents. With that in mind, this study was focused on developing an AI-based littering detection system using the YOLOv5 algorithm for barangays which will help promote discipline towards the people within the vicinity when it comes to refraining themselves from dumping trash as the CCTVs can catch them during such an act and send an SMS alert notification to the assigned barangay official. In that way, violators will be easily detected, and the personnel in charge can penalize them for being irresponsible with their trash. RESEARCH METHODOLOGY: The study implemented an experimental research design and utilized the Agile methodology of the System Development Life Cycle (SDLC) for the development of software and hardware. In developing the littering detection system, a custom dataset was created in the Roboflow application. The generated dataset was trained using the YOLOv5 algorithm in Pycharm through the use of Phyton programming language. On the other hand, the barangays Graphical User Interface (GUI) of the system was developed utilizing the C# programming language in Visual Studio 2022. The constructed littering detection model was embedded in the Rasberry Pi 4 Model B. Additionally, the GSM Module SIM800L v2 was used to send an SMS notification when littering is detected. The system underwent multiple tests, which included littering detection accuracy testing. SMS alert notification testing, unit testing, and system testing to evaluate the functionalities of the Dalis-AI. SUMMARY OF FINDINGS: The YOLOv5n model with 1280 image pixels was selected as the model for the littering detection system because it was the most reliable detection with a computed average F1-score of 81.5%, an mAP of 88.2%, and an inference speed of 1-2 seconds. For each instance of detecting littering, the SMS alert notification system was able to send a message that contained the system’s name, position, date, and time. Based on the Dalis-AI’s two-day testing, the researchers validated a total of 23 actual littering violation and had an average success rate of 78.26% for littering detection with five false positives (FNs). The experiment revealed that the Dalis-AI could detect littering and always notify the authorized barangay authority an SMS alert. On the other hand, Dalis-AI is sometimes unable to detect individuals who drop too small pieces of trash like candy wrappers and cigarette butts. Additionally, the system’s detection can be compromised by the lighting and the distance of the violators from the camera. CONCLUSION: The YOLOv5 algorithm was successfully utilized in the project to create an AI CCTV integrated littering detection system for Barangay 439, Sampaloc, Manila. The Dalis-AI has the capacity to detect instances of littering and enable quick action by sending an SMS alert notification in real time. The SMS alert notification system ensures that barangay officials are notified immediately, enabling them to address the littering violation swiftly and effectively. The Dalis-AI’s functionalities were effectively tested and evaluated. The Dalis-AI can be considered to be fully implemented in their vicinity to mitigate improper waste disposal. RECOMMENDATION: In this study, several recommendations may be made to improve future studies. The latest version of the YOLO algorithm can be considered for future research. The custom dataset can be enhanced by adding more footage of individuals throwing out different kinds and sizes of trash in order to produce a model that is more accurate. Given that the NVIDIA Jetson has a built-in GPU and accelerates detection, future developers may utilize it as a microprocessor. The Dalis-AI was only paralleled to the barangay’s current CCTV because of limited storage and financial constraints. The researchers recommend using a larger memory storage and giving the barangays GUI the capability to record the live stream that was captured by the IP camera for it to function as a standalone system. In order for the system to have proper ventilation and to blend in wherever it is placed in barangays, the study recommends designing a custom, seamless, and compact casing. Moreover, future researchers can replicate the concept of the study with a different objective based on the barangay’s primary concern.
academic writing
TK8360 C33 2023
Dalis-AI: An AICCTV integrated littering detection system using yolov5 algorithm for Barangay 439, Sampaloc, Manila
ABSTRACT: STATEMENT OF THE PROBLEM: One of the causes of the widespread garbage problem is improper waste disposal which starts in small-scale communities called barangays. It can be observed that most people often throw their garbage in the streets. Some residents are unaware of the consequences of their actions and continue not to follow the rules regarding proper waste disposal. Aside from that, because of the lack of discipline of the residents, littering results in adverse effects on the community, such as clogging of drainages, unpleasant scenery, and harmful effects on the health of the residents. With that in mind, this study was focused on developing an AI-based littering detection system using the YOLOv5 algorithm for barangays which will help promote discipline towards the people within the vicinity when it comes to refraining themselves from dumping trash as the CCTVs can catch them during such an act and send an SMS alert notification to the assigned barangay official. In that way, violators will be easily detected, and the personnel in charge can penalize them for being irresponsible with their trash. RESEARCH METHODOLOGY: The study implemented an experimental research design and utilized the Agile methodology of the System Development Life Cycle (SDLC) for the development of software and hardware. In developing the littering detection system, a custom dataset was created in the Roboflow application. The generated dataset was trained using the YOLOv5 algorithm in Pycharm through the use of Phyton programming language. On the other hand, the barangays Graphical User Interface (GUI) of the system was developed utilizing the C# programming language in Visual Studio 2022. The constructed littering detection model was embedded in the Rasberry Pi 4 Model B. Additionally, the GSM Module SIM800L v2 was used to send an SMS notification when littering is detected. The system underwent multiple tests, which included littering detection accuracy testing. SMS alert notification testing, unit testing, and system testing to evaluate the functionalities of the Dalis-AI. SUMMARY OF FINDINGS: The YOLOv5n model with 1280 image pixels was selected as the model for the littering detection system because it was the most reliable detection with a computed average F1-score of 81.5%, an mAP of 88.2%, and an inference speed of 1-2 seconds. For each instance of detecting littering, the SMS alert notification system was able to send a message that contained the system’s name, position, date, and time. Based on the Dalis-AI’s two-day testing, the researchers validated a total of 23 actual littering violation and had an average success rate of 78.26% for littering detection with five false positives (FNs). The experiment revealed that the Dalis-AI could detect littering and always notify the authorized barangay authority an SMS alert. On the other hand, Dalis-AI is sometimes unable to detect individuals who drop too small pieces of trash like candy wrappers and cigarette butts. Additionally, the system’s detection can be compromised by the lighting and the distance of the violators from the camera. CONCLUSION: The YOLOv5 algorithm was successfully utilized in the project to create an AI CCTV integrated littering detection system for Barangay 439, Sampaloc, Manila. The Dalis-AI has the capacity to detect instances of littering and enable quick action by sending an SMS alert notification in real time. The SMS alert notification system ensures that barangay officials are notified immediately, enabling them to address the littering violation swiftly and effectively. The Dalis-AI’s functionalities were effectively tested and evaluated. The Dalis-AI can be considered to be fully implemented in their vicinity to mitigate improper waste disposal. RECOMMENDATION: In this study, several recommendations may be made to improve future studies. The latest version of the YOLO algorithm can be considered for future research. The custom dataset can be enhanced by adding more footage of individuals throwing out different kinds and sizes of trash in order to produce a model that is more accurate. Given that the NVIDIA Jetson has a built-in GPU and accelerates detection, future developers may utilize it as a microprocessor. The Dalis-AI was only paralleled to the barangay’s current CCTV because of limited storage and financial constraints. The researchers recommend using a larger memory storage and giving the barangays GUI the capability to record the live stream that was captured by the IP camera for it to function as a standalone system. In order for the system to have proper ventilation and to blend in wherever it is placed in barangays, the study recommends designing a custom, seamless, and compact casing. Moreover, future researchers can replicate the concept of the study with a different objective based on the barangay’s primary concern.
academic writing
TK8360 C33 2023