Abstract: At present, the result of the 3D pavement crack detection method is bad. This paper proposes a new pavement crack extraction method based on Hessian matrix and linear structure. Firstly, in order to solve the problem of weak contrast and multi interference of pavement depth image, according to the characteristics of edge and gray level of crack, a new algorithm based on piecewise function is proposed. Then, in view of the issue of discontinuous crack and sensitivity to noise of multi scale rough extraction algorithm based on Hessian matrix, according to the consistency of the local direction of crack and the discreteness of noise, a new algorithm based on linear structure constraint is proposed. The experimental results show that the method is suitable for pavement cracks with different scales and degrees, compared with the method of document［5］ and document［15］, the extraction integrity is improved by more than 10% and more than 6% respectively, the extraction correct rate is increased by more than 2% and more than 1% respectively, but the time is almost the same as the first and is about one third of the second.

Key words: depth image, piecewise function, crack extraction, linear structure