Drone object detection and tracking.
I've created my own dataset for a remote controlled helicopter.
Used image augmentation in order to enhance the dataset.
Training dataset consists of frames extracted from videos recorded both inside and outside
The most surprising result can be seen in the last picture, in which the network detects the helicopter shadow.
Movement commands are issued based on how far off the tracked object is from the center of the image. The farther away, the more aggressive the movement will be.
For tracking I've used a pretrained network (VOC) as controlling the helicopter and making sure the drone wouldn't crash at the same time would have been a challenge.
Spaghetti code, feel free to ask anything.
https://github.com/weiliu89/caffe/tree/ssd
https://github.com/chuanqi305/MobileNet-SSD
https://github.com/srianant/kalman_filter_multi_object_tracking
https://github.com/brean/python-ardrone