Segnet

http://mi.eng.cam.ac.uk/projects/segnet/ A Deep Convolutional Encoder-Decoder Architecture for Robust Semantic Pixel-Wise Labelling Use a random image, upload your own, search for a place, or click on one of the example images in the gallery below. SegNet is trained to lassify each pixel of an urban street image to be one of twelve classes: pedestrian, car, building, tar road, trees etc.


 * https://github.com/alexgkendall/caffe-segnet.


 * See Caffe berkeley vision

Kai Yan
http://diydrones.com/profile/yankai  implements on Jetson tegra(nvidia),  Cctv_cameras(https://www.e-consystems.com/See3CAM-80.asp)

http://diydrones.com/profiles/blogs/control-a-copter-by-image-recognition?xg_source=activity