In rolling shutter (RS)-based optical camera communication (OCC) links, selecting the appropriate camera's exposure time is critical, as it limits the reception bandwidth. In long exposures, the pixels accumulate over time the incoming irradiance of several consecutive symbols. As a result, a harmful intersymbol interference corrupts the received signal. Consequently, reducing the exposure time is required to increase the reception bandwidth at the cost of producing dark images with impracticable light conditions for human or machine-supervised applications. Alternatively, deep learning (DL) equalizers can be trained to mitigate the exposure-related ISI. These equalizers must be trained considering the transmitter clock and the camera's exposure, which can be exceptionally challenging if those parameters are unknown in advance (e.g., if the camera does not reveal its internal settings). In those cases, the receiver must estimate those parameters directly from the images, which are severely distorted by the exposure time. This work proposes a DL estimator for this purpose, which is trained using synthetic images generated for thousands of representative cases. This estimator enables the receiver operation under multiple possible configurations, regardless of the camera used. The results obtained during the validation, using more than 7000 real images, registered relative errors lower than 1% and 2% when estimating the transmitter clock and the exposure time, respectively. The obtained errors guarantee the optimal performance of the following equalization and decoding receiver stages, keeping bit error rates below the forward error correction limit. This estimator is a central component of any OCC receiver that operates over moderate exposure conditions. It decouples the reception routines from the cameras used, ultimately enabling cloud-based receiver architectures.