Front crawl swimming is a cyclic activity that can be broken down into stroke phases. Traditionally, video analysis is considered as the gold standard technique to identify these key points of the cycle (entry, pull, push, and recovery) but is very time-consuming and somewhat error-prone. The development of embedded sensors, such as inertial measurement units (IMUs) combining 3-D accelerometer, 3-D gyroscope, and 3-D magnetometer, offer possibilities to automatically detect such points. Our study aimed to develop and validate a new algorithm that would adapt to several constraints (various swimming speeds and a variety of swimming kinematics for para swimmers with physical impairments) in order to accurately detect the beginning of each stroke phase. Two IMUs positioned on the dorsal sides of the lower arms were used in this study. Linear regressions, Bland and Altman diagrams, and root-mean-square errors (RMSEs) highlighted the agreement and accuracy of the algorithm in comparison to traditional video analysis. The absence of a plateau for the detection of the pull at fast speeds or spatiotemporal changes in the signal of angular velocity for para swimmers were all examples of the high degree of adaptability of the developed algorithm. Therefore, no significant difference was observed between the two techniques and for all conditions (swimming at slow or fast speeds for non-disabled or para swimmers) for the detection of any of the four key points of the front crawl stroke. Such results offered a possibility to quantify accurately the degree of asymmetry of the upper limb coordination for swimmers with physical impairments. We revealed that the para swimmers were able to dissociate their coordination by reducing the lag time between two propulsive actions of the limbs on one body side, to compensate for the absence or limited action of the impaired limb on the controlateral side. Practically, such methodological improvements should be integrated by coaches and swimmers in the development of the monitoring of their training activities to accurately quantify how the swimmers adapt to their different nature surrounding constraints.