A Generalized Multi-Bernoulli Filter for Track-before-detect (GMB-TBD) of objects from image observations when the objects’ influence region overlapping is proposed. The overlapping objects’ measurement likelihood function is analyzed, the likelihood function is estimated by predicted objects’ states, and then objects’ overlapping influence is eliminated on objects’ states updating by using this estimation. In this filter, the predicted and updated objects’ states are strictly assumed as Multi-Bernoulli RFS, so it’s a true Multi-Bernoulli based TBD filter and it can be used under both the objects’ influence region overlapping and non-overlapping situations. The filter’s realization steps are given. Objects’ tracks are pruned and extracted by labeling Multi-Bernoulli components. Lastly, GMB-TBD filter’s performance is verified by computer Monte-Carlo simulation results.