This module provides the APIs to initialize, query, and run the NVIDIA proprietary parking space detection neural network: ParkNet. The included data structures are as following:
dwParkNetModel, which allows users to choose from a default or user-trained ParkNet model.dwParkNetParams, which defines the ParkNet model variant with specific precision to be loaded.ParkNet consumes planar frames in FP16 precision with a resolution of 960x604 pixels from AR0231 cameras. ParkNet only processes the bottom 960x480 pixels; cropping is handled in the data-conditioner code. ParkNet has been trained with images from cameras forming a surround coverage around the vehicle. Individual cameras from the surround-setup had a 60° or 120° field of view. Assume that in inference mode, ParkNet is applied to images of similar nature.
ParkNet detects multiple available parking spaces in image-space. Each detected parking space is also provided with an identified entry line and a confidence of the detection. The actual processed output from the ParkNet DNN is presented in a form of a dwParkDetection structure by the ParkNet detector module described in Parking Space Perception.
0.45 is applied for all parking spaces confidences to determine if a parking space should be returned from the detector. This can result in the parking spaces output flickering if these confidences are oscillating around the threshold value.