The construction of an appropriate observation matrix is the key issue for compressed sensing practical application in embedded system. However, the matrix usually used, like Gaussian matrix, Bernoulli matrix etc, is difficult to realize for hardware. Aiming at the feasibility and the real-time of compressive sensing for embedded vision system, a simple deterministic 0-1 observation matrix is proposed, which is based on the related work of sparse transform vectorization, the characteristic of wavelet sparsity, and the pseudo-random sequence observation matrix construction. For N×N image, when the observation matrix dimension is M×N, the matrix is consist of M base vectors and the vector size is N. The base vector has only one element of 1 and the other elements are 0. So the formed observation matrix is simple and the integral circuit can be omitted in the real measurement. Base vectors are arranged according to the position of element 1 in vector from 1 to M to form a deterministic observation matrix, which is easy to remember and store the matrix, and also helpful for reconstruction.