Crowdsourcing systems connect task publishers with a large, diverse pool of workers, but traditional centralized platforms raise concerns on data integrity, trust, and fair payment. While blockchain-based crowdsourcing offers decentralization and transparency, it also creates opportunities for collusion between blockchain nodes and system users, thereby undermining system reliability. Existing collusion-resistant methods cannot be directly integrated into decentralized settings and generally overlook collusion targeting incentive information. To address these limitations, we propose TriDA, a triangular fuzzy number-based double auction framework, to prevent incentive information leakage through collusion and to achieve efficient and accurate task-worker matching and reward determination. Specifically, we classify incentive information into functional (e.g., deadline and price) and eligibility (e.g., maximum geographic distance, minimum label matching degree) types that are modeled using triangular fuzzy numbers. We design TriMatch, an algorithm to filter infeasible task-worker pairs, compute parameter-wise matching degrees considering both parties’ perception of the matching pairs and their risk preferences, and aggregate them into an overall matching degree to maximize social welfare. The workers’ rewards are subsequently determined via an apex-overlap method, ensuring fairness and interpretability. Extensive experimental results demonstrate that TriDA achieves high matching efficiency and effectively reduces matching deviations compared to existing methods.