Artificial intelligence inference on protected media content in a vision processing unit

A network video recorder system comprising:

• a host to receive an encrypted bitstream from one of a plurality of cameras;
• and
• a vision processing unit coupled to the host via only a memory buffer of the vision processing unit, the vision processing unit further comprising:
• a security engine to decrypt the encrypted bitstream based on a digital rights management key corresponding thereto;
• a video decode processor to decode the decrypted bitstream to generate video comprising at least one video frame, the security engine to encrypt the video based on a local key and to store the encrypted video to the memory buffer;
• and
• a neural network processor to retrieve the encrypted video from the memory buffer, decrypt the encrypted video via a direct memory access to the security engine and to apply a neural network to the decrypted video to generate one or more artificial intelligence parameters corresponding to the at least one video frame.

A method for performing artificial intelligence on protected video content comprising:

• receiving, via a vision processing unit, an encrypted bitstream representative of protected video content;
• decrypting the encrypted bitstream based on a digital rights management key corresponding thereto, decoding the decrypted bitstream to generate video comprising at least one video frame, encrypting the video based on a local key, and storing the encrypted video to memory;
• and
• retrieving the encrypted video from the memory, decrypting the encrypted video via a neural network processor of the vision processing unit based on the local key, and applying a neural network via the neural network processor to the decrypted video to generate one or more artificial intelligence parameters corresponding to the at least one video frame.