AbstractCamera traps are a powerful research tool with a wide range of applications in animal ecology, conservation, and management. However, camera traps may not always detect animals passing in front, and the probability of successfully detecting animals (i.e. camera sensitivity) may vary spatially and temporarily. This constraint may create a substantial bias in estimating critical parameters, such as the density of unmarked populations or animal activity levels.We applied the ‘double-observer approach’ to estimate detection probability and correct potentially imperfect detection. This involved two camera traps being set up at a camera station to monitor the same focal area. The detection probability and the number of animal passes were concurrently estimated with a hierarchal capture-recapture model for stratified populations using a Bayesian framework. Monte Carlo simulations were performed to test the reliability. We then estimated the detection probabilities of a camera model (Browning Strike Force Pro) within an equilateral-triangle focal area (1.56 m2) for 12 ground-dwelling mammals in Japan and Cameroon. We also evaluated the possible difference in detection probability between daytime and nighttime by incorporating it as a covariate.We found that the double-observer approach reliably quantifies camera sensitivity and provides unbiased estimates of the number of animal passes, even when the detection probability varies among animal passes or camera stations. The camera sensitivity did not change between daytime and nighttime either in Japan or Cameroon, providing the first evidence that the number of animal passes per unit time may be a viable index of animal activity levels. Nonetheless, the camera traps missed animals within the focal area by 4 %–36%. Current density estimation models relying on perfect detection may underestimate animal density by the same order of magnitude.Our results showed that the double-observer approach might be effective in correcting imperfect camera sensitivity. The hierarchical capture-recapture model used here can estimate the distribution of detection probability and the number of animals passing concurrently, and thus, it is easily incorporated in the current density estimation models. We believe that this approach could make a wide range of camera-trapping studies more accurate.