Intelligent Second Look, or I2L, required the deployment of custom automated detection in high resolution live video on a resource constrained device. Deep Analytics was able to provide AI/ML capability with excellent performance in a 4K streaming and 52MPix capturing camera system (built in-house) by developing custom processor and memory handling routines.

Specifically, the 52MPix image formation, originally a serialized script, was instead rewritten to execute across 6 processing threads in parallel. Similarly, the onboard live video server was parallelized. Various functions including button presses, LED indicators, and LED illuminators were offloaded onto an embedded microcontroller.

Additional processing overhead made available with these optimizations enabled the execution of AI/ML classifier in live video, but various camera functions were limited by memory. To solve the remaining issue, DA developed a driver to dynamically allocate CPU and GPU memory, effectively ensuring a task-specific resource table for any given camera function. Altogether, the lightweight custom Linux distribution, supported by a microcontroller, provided a stable environment to perform advanced real-time camera capabilities and AI/ML functionality.

View: All Case Studies | OUR Design EXPERTISE