Design and Analysis of Video Desensitization Algorithm Based on Lightweight Model PP PicoDet

Intelligent connected vehicle data has multiple application scenarios and a large volume of data, and the security issues of connected vehicle data, especially the issue of illegal collection of faces and license plates outside the vehicle, are becoming increasingly prominent. According to the latest regulatory requirements, intelligent connected vehicles need to be desensitized before transmitting facial and license plate images outside the vehicle. However. Due to the special requirements for memory resources and performance, deploying desensitization algorithms on the vehicle side faces a huge challenge and pressure. The PP PicoDet model is a lightweight object detection model that utilizes carefully designed network architecture and optimization strategies to ensure high detection accuracy while reducing the model's parameter and computational complexity. This makes the deployment of PP-PicoDet on onboard devices more efficient and feasible. To address this prominent industry issue, this article adopts the lightweight model PP-PicoDet to reduce the algorithm's demand for on-board equipment resources, thereby reducing the pressure of desensitization algorithms in vehicle deployment. Experimental analysis shows that this algorithm can achieve a face and license plate detection rate of >96%, a false detection rate of <5%, and further reduce resource consumption. This is an important technological breakthrough for the intelligent connected vehicle industry, which helps to promote the development of data security in the Internet of Vehicles.