As a new ferroelastic state, strain glass has attracted a lot of recent attentions and, most importantly, strain glass transitions (SGTs) could underpin many phenomena that have puzzled the physics community for decades, including the quasi-linear superelasticity and Invar and Elinvar anomalies. However, there has been a lack of fundamental understanding at the atomistic level beyond the phenomenological Landau theory. In this paper, we propose a way to obtain quantitatively the continuous strain/stress fields distribution caused by point defects through molecular statics calculations by incorporating a Gaussian probability distribution function. By using the quantitative strain/stress fields distribution to inform phase field simulations, we reproduce quantitatively the experimentally observed critical defect concentrations separating the normal martensitic phase transition from SGTs at different temperatures and critical temperatures for spontaneous strain glass to martensitic transition at different defect concentrations. Based on percolation theory, we demonstrate how the strain network created by point defects with a critical concentration regulates the nucleation and growth of martensitic domains, suppresses autocatalysis by strain frustration, and changes the sharp first-order martensitic transformation into a continuous SGT. A general temperature- and defect-concentration-dependent percolation criterion is formulated for accurate prediction of SGT, which could enable high throughput computations for systematic search of new strain glass systems using simply molecular static calculations.