Recreational fisheries often pose various challenges for monitoring. Conventional techniques, such as aerial-roving creel surveys, can lead to estimation bias as they are limited to sampling fragments of these fisheries, which are often fluid and temporally dynamic. In British Columbia, Canada, the existing aerial-roving framework for monitoring marine recreational salmon fisheries relies on angler testimony to report diel temporal distribution of angling and flights to scale-up effort estimation. This work involved implementing two shore-based autonomous cameras set at high capture-rates. One system allowed for the estimation of harbour-specific effort and retention, while the second system informed the timing of angler use by monitoring a popular nearby fishing location. We employed a two-step image filtering system which included the use of a novel deep learning-based system involving a Gaussian Mixture Model-based motion detector for the automatic detection of vessels, to efficiently analyze over one million images. We calculated effort and retention estimates using catch-per-unit effort data from interviews and detailed the diel activity of observed fishing using empirical cumulative distribution and activity pattern analyses. We found that creel estimates were predicting the timing of fishing relatively well, but alone, they may be failing to account for substantial effort in the afternoons. We also provided new insight into a notable amount of off-season fishing effort, which is currently lacking in creel estimates, by surveying beyond peak-season. Our findings suggest that cameras can provide a viable tool in British Columbia for extending the temporal resolution of existing methods while providing an efficient and cost-effective alternative to flights to expand effort data. Increasing camera monitoring to cover additional marinas and popular fishing locales could improve estimation and protect vital fish species.