Meshine

Swarm Robotics

Swarm robotics aims to perform the tasks which robots will perform individually at high cost in a long time or the tasks they cannot be complete by a "swarm" consisting of many robots. As Meshine, we are developing systems that target unmanned aerial vehicles to perform complicated tasks using these swarm algorithms. In these systems, in addition to having some abilities individually, each member of the swarm gains abilities such as acting collectively, recognizing the detected object by the whole swarm and making common task distribution accordingly. These systems can be used in many industries, especially security, entertainment, agriculture, energy.


Object Detection

Hac facilisi ac vitae consec tetu commodo vel magna suspendisse on senectus pharetra magnfauc


Collective Motion

Hac facilisi ac vitae consec tetu commodo vel magna suspendisse on senectus pharetra magnfauc


Object Tracking

Hac facilisi ac vitae consec tetu commodo vel magna suspendisse on senectus pharetra magnfauc


Human Detection and Tracking with UAV images

In our study, at first stage, image data at certain heights were collected from the ground. Defined people in images. These data are trained with deep neural networks. Learning models have been developed to enable rapid identification. As a result, detection and tracking has been achieved with more than 70% consistency in human images taken from 3 different recordings. Light electronic development boards have been preferred as much as possible to optimize the use of the application in mini UAV. In this case, the decrease in FPS values ​​is taken into account in order to keep the processing power at the proper level.

Simultaneous Object Detection and Path Planning with UAVs

In this project, work was carried out on both path planning and object detection using more than one neural network model. In this way, the UAV was provided to avoidance from obstacles while in autonomous movement. It is also planned to detect threats around it. Our first artificial neural network is focused on depth estimation operations normally done by binoculars or dual cameras. However, thanks to the innovative work carried out, monocular, that is, has been trained to produce a depth map with a single camera. The first use of this technology used for autonomous cars in Germany for a drone was realized with this project. In this way, with lower cost Consistent distance measurements were obtained from the image. The second neural network focused on the detection of objects such as weapons and knives that could be a threat. Another function is to add endemic plants to our network by adding the data used for endemic plant determinations. A simulation environment has been created through the game engine in order to carry out the test stages safely and quickly. In this simulation environment, the drone follows the previously planned route thanks to the autopilot software we have. On the one hand, it has managed to avoid obstacles in the artificial environment and on the other hand it has successfully performed plant and weapon detection operations.

Aflotoxin Detection on Dried Foods Using Image Proccessing

Aflatoxin is toxic waste produced by a group of fungus. It is mostly seen on dried foods like fig, hazelnut and pistachio. Our system detects the products and aflatoxin area on products using image processing. This system can be integrated on variaty of industrial systems like sorting machines and industrial microwave ovens.