This paper proposes an adaptive neural network finite time command filtered control scheme based on observers for a class of nonlinear systems with full state constraints and unmeasured states. Firstly, the integral barrier Lyapunov functions are constructed to guarantee that the state variables satisfy the time-varying constraint conditions. Secondly, considering that the states of most nonlinear systems are unmeasurable, an adaptive neural network state observer is designed to estimate the unmeasurable states. Then, the finite time command filtered control method is used in the process of controller design to avoid the phenomenon of ‘differential explosion’, and the error compensation mechanism is further introduced to eliminate the filtering errors. Besides, the dead zone model is described as the form of linear input and bounded disturbance to solve the input dead zone problem. The proposed method ensures that the system states do not exceed the constraint bounds, all signals in the closed-loop system are bounded in finite time, and the tracking error can converge to within the neighborhood near zero in finite time. Finally, the effectiveness of the proposed method is verified by two examples.