How to train a neuron to recognize objects in a video?

Question

Neural networks

People orient plz. It is necessary to train the neuron on the video recording to identify objects (including the acceleration of these objects, the distance traveled, etc.). You need a sequence of actions and materials on the topic.
1. For this purpose, which neuron should be taken as a basis (from Google, Microsoft, etc.)?
2. In what language is it better to teach her?
3. What algorithm to use?
4. How to prepare data, etc.?
5. Perhaps there is a certain Framework, a wrapper for a neuron?

Answer 1

If the definition of objects means detection and recognition, then:
1. You can take the detection api from tensorflow (python, there seems to be an api for c ++) or YOLOv2 (c? c ++, python).
2. Neurons work with video as a sequence of images => for tracking and everything else, data from neurons must be transferred to the tracker (you can use the Hungarian algorithm), google in the direction of multi object tracking.