Tracking Multiple Objects Using Computer Vision and Inertial Measurement Unit
Björklund, Jani (2021)
2021
Master's Programme in Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2021-04-28
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202104273774
https://urn.fi/URN:NBN:fi:tuni-202104273774
Tiivistelmä
Industrial areas are vast and have massive movement going in and out. Knowing who is in the area and where they are approximately located is one of the main goals in access control. To achieve this goal, the industrial area has to have dedicated infrastructure for it. The problem arises from the expensive investments and lack of easy scalability.
This work describes a prototype system’s design and research based on computer vision and mobile devices that track people in a restricted industrial area. Using computer vision with IP cameras is more affordable to cover bigger areas and easier to scale the infrastructure to match the needs.
The prototype system consists of two IP cameras that send video surveillance feed to the server to be analyzed and mobile devices that send sensor data from a user. The implemented system consists of two Xiaomi Redmi Note 8T phones as IP cameras that send surveillance feed wirelessly to Python-server to be processed. The server’s convolutional neural networks recognize a person from the image and aim to re-identify the person during the whole visit in the area. The person’s movement is followed on a map, where the location on the image is projected upon. If the person is not detected from the image, the location’s estimation is sought using the same person’s mobile device. By fusing these two tools, a more robust tracking system is created. Besides these tools, also an option to use satellite tracking in the system was researched.
The system was tested by finding out if the system can operate well in the areas where the satellite tracking was having issues and if the satellite tracking can identify a person seen by the system’s cameras. Besides comparing the proposed system to the satellite tracking, the tracking visually and dead reckoning was analyzed. Finally, the system’s capability to follow multiple people was tested.
The testing results indicate that it is possible to track multiple objects accurately by using visual detections. Besides visual tracking, it is also possible to get an estimated position based on dead reckoning when the object is not visually detected. The satellite tracking cannot follow the person properly in challenging environments, but it can be used to identify the person in the open areas.
This work describes a prototype system’s design and research based on computer vision and mobile devices that track people in a restricted industrial area. Using computer vision with IP cameras is more affordable to cover bigger areas and easier to scale the infrastructure to match the needs.
The prototype system consists of two IP cameras that send video surveillance feed to the server to be analyzed and mobile devices that send sensor data from a user. The implemented system consists of two Xiaomi Redmi Note 8T phones as IP cameras that send surveillance feed wirelessly to Python-server to be processed. The server’s convolutional neural networks recognize a person from the image and aim to re-identify the person during the whole visit in the area. The person’s movement is followed on a map, where the location on the image is projected upon. If the person is not detected from the image, the location’s estimation is sought using the same person’s mobile device. By fusing these two tools, a more robust tracking system is created. Besides these tools, also an option to use satellite tracking in the system was researched.
The system was tested by finding out if the system can operate well in the areas where the satellite tracking was having issues and if the satellite tracking can identify a person seen by the system’s cameras. Besides comparing the proposed system to the satellite tracking, the tracking visually and dead reckoning was analyzed. Finally, the system’s capability to follow multiple people was tested.
The testing results indicate that it is possible to track multiple objects accurately by using visual detections. Besides visual tracking, it is also possible to get an estimated position based on dead reckoning when the object is not visually detected. The satellite tracking cannot follow the person properly in challenging environments, but it can be used to identify the person in the open areas.