Pedestrian behavior has much more complicated characteristics in a dense crowd and thus attracts the widespread interest of scientists and engineers. However, even successful modeling approaches such as pedestrian models based on particle systems are still not fully considered the perceptive mechanism underlying collective pedestrian behavior. This paper extends a behavioral heuristics-based pedestrian model to an adaptive agent-based model, which explicitly considers the crowding effect of neighboring individuals and perception anisotropy on the representation of a pedestrians visual information. The adaptive agents with crowding perception are constructed to investigate complex, selforganized collective dynamics of pedestrian motion. The proposed model simulates selforganized pedestrian flows in good quantitative agreement with empirical data. The selforganized phenomena include lane formation in bidirectional flow and fundamental diagrams of unidirectional flow. Simulation results show that the emergence of lane formation in bidirectional flow can be well reproduced. To investigate this further, increasing view distance has a significant effect on reducing the number of lanes, increasing lane width, and stabilizing the self-organized lanes. The paper also discusses phase transitions of fundamental diagrams of pedestrian crowds with unidirectional flow. It is found that the heterogeneity of how pedestrians perceive crowding in the population has a remarkable impact on the flow quality, which results in the buildup of congestion and rapidly decreases the efficiency of pedestrian flows. It also indicates that the concept of heterogeneity may be used to explain the instability of phase transitions.