Entropy harvest and key derivation from the image sensors in IP camera

With the widespread use of connected surveillance systems using IP cameras, the security of video data has become a critical issue. In response to the potential compromise of sensitive information via covert channels, the United States and the United Kingdom warned against the use of suspicious cameras in critical infrastructure. To mitigate the security concerns, it is indispensable to investigate every component inside IP cameras using evaluation and validation programs such as Common Criteria and Cryptographic Module Validation Program. However, such compliance tests on implementation under test cannot suffice to prevent the products from malicious attacks using a Trojan-horse in a black-box component, particularly inserted in the image-sensing module or the random number generator (RNG). In this paper, we categorize potential vulnerabilities and argue that IP cameras should disclose the components of the modules to mitigate the threats of backdoor. Based on this, we provide a way to construct backdoor-free RNG. To remove the possibility of the backdoor inside the RNG, we extract randomness from optical black pixels of the image sensor and apply two entropy accumulation methods to guarantee the min-entropy of accumulated data under weak assumptions, which means the IP camera already has a hardware entropy source in itself without additional suspicious entropy source. In the experiment using an IP camera with a Cortex-A53 processor, we can harvest entropy at the rate of 7.3kbps. The operating speed of the RNG is sufficient to provide random bits without delay when a symmetric key is updated every minute.