Revision history [back]

Well it kind of depends, but my best guess would be to go for a standard object detection model trained on a dataset that already contains many of the general classes on the street, like cars, trucks, pedestrians, ... have a look at

• The model zoo supplied by Intel https://github.com/opencv/open_model_zoo/tree/2018/intel_models
• These models can be read in by the DL backend: https://github.com/opencv/opencv/tree/master/modules/dnn
• There are a bunch of usefull tutorials: https://docs.opencv.org/master/d2/d58/tutorial_table_of_content_dnn.html