How to approach camera tuning
Tuning is a continuous process, so its important to start from a spec and finish it when the spec is met. Spec is compiled by modulating optical benchmarking results with perception requirements. First reference point is created by benchmarking available solution. Next spec is being constructed by translating perceptual requirements into optical and HW specs.
Perceptual requirements are collected by researching vision trend and categorization of camera issues.
Read about vision trend research in the following article
In robotic and OEM companies issues are categorized by frequency and degree of malfunctioning of the agent and in applications targeted for human vision - by surveying the target group.
Perception requirements can be as simple as removing color tint or more advanced like being able to detect a certain moving object at certain distance, speed and illumination. In the spec its important to register all operational states and be able to classify whether camera is optimized for detection of only certain type of objects or needs to equally treat everything that we observe in the scene. If the agent knows its state and target and can communicate it to the ISP we can dynamically change configuration.
Tuning hyperloop is cycling through iterations of benchmarking and tuning sessions unless the performance spec is met.
