Abstract: Objectives: Due to its strong flexibility, low cost, and continuous availability, dockless shared bicycles have become the primary form of bike-sharing services in China. However, the excessive flexibility of dockless shared bicycles often leads to a supplydemand imbalance, manifested by scarcity within specific temporal and spatial domains, where bikes cannot be found or accumulate without being used. Accurately and precisely detecting the spatial and temporal extent, as well as the degree of imbalance in the origin-destination (OD) distribution of shared bicycle trips is a crucial issue for understanding the reliable cognition of bike supply and demand relationships, scientifically dispatching bike resources, and developing green and sustainable public transportation. Currently, most studies use predefined grids to divide spatial units and then construct metrics to measure the degree of OD imbalance, setting imbalance thresholds to extract source and sink regions. The method of manually predefining rigid spatial units and setting imbalance thresholds tends to underestimate the spatial and temporal extent of sources and sinks and the degree of imbalance. Additionally, there is a lack of coupled analysis of the temporal and spatial evolution of travel source and sink areas. Methods: To this end, this paper proposes a bivariate clustering method based on spatiotemporal density to adaptively extract trip sources and sinks. Firstly, spatial kernel density estimation and information entropy theory are introduced to adaptively extract OD core points from the perspective of density significance test. Secondly, a bivariate distribution imbalance statistical method is constructed to identify source and sink core points of shared bike trips, and density-based clustering extension strategies are implemented for fine-grained identification of source-sink spatial regions. Finally, based on the detected source-sink spatial regions, OD point event intensity and imbalance significance tests are designed based on OD point temporal density distribution and event type probability distribution, to identify the sources and sink time periods of shared bike trips. Results: Experimental results using dockless shared bicycle trajectory data in Xiamen Island reveal that within the morning peak hours, this method identified 11,351 source areas and 9,453 sink areas. The number of sources and sinks can reach up to nearly 2,000 per hour. The imbalance index for source areas is concentrated in the range 0.5, 1, while for sink areas, it is concentrated in the range -1, -0.5. This demonstrates that the proposed method can accurately and reliably extract spatial regions, durations, and degrees of imbalance that characterize the distribution imbalance of vehicle borrowing and returning. Conclusions: In order to overcome the problems that the spatial detection results of shared bicycle travel source and sink are heavily dependent on the rigid pre-division of spatio-temporal units, the predefined threshold of algorithm parameters and source and sink indicators, and the lack of spatio-temporal coupling mining analysis of source and sink, this paper proposes an adaptive spatial detection algorithm for traffic travel source and sink based on binary variable density clustering. A case study was carried out in the actual data of shared bicycles in Xiamen.