Objective Targeting at the movement features of bags on conveyer in industrial scene, we proposed an intelligent image counting method for the online moving bags. Methods Firstly, the probability distributions of the number of '1'of the foreground image were computed, and then the probability distribution was approximated to the continuous probability distributions. In order to calculate the probability distribution density S（t） , the linear difference was used to calculate a time probability distribution of the foreground image. Finally, the S（t）values were compared and the moving bag counting was achieved. Results The packet loss problem during bag statistics was solved, and the accuracy of bag counting was significantly improved. The error detection rate dropped to 1%. Conclusion The method has the advantage of high counting accuracy, and it overcomes the shortcomings of traditional algorithms. The accuracy of the counting has been effectively improved.