Construction workers' unsafe behavior is one of the main reasons leading to construction accidents. However, the existing management approach to unsafe behaviors, e.g. Behavior-Based Safety (BBS), relies primarily on manual observation and recording, which not only consumes lots of time and cost but impossibly cover a whole construction site or all workers. To solve this problem and improve safety performance, an image-skeleton-based parameterized method has been proposed in a previous research to real-time identifying construction workers' unsafe behaviors. A theoretical framework has been developed with a preliminary test, but still lacking a comprehensive experiment to verify its validity, particularly in the recognition of the types of unsafe behaviors. This will have a serious impact on the extensive application of the method in real construction sites. Based on the method, this research designs and carries out a series of experiments involving three types of unsafe behaviors to examine its feasibility and accuracy, and determines the value ranges of relevant key parameters. The results of the experiment demonstrate the feasibility and efficiency of the method, being able to identify and distinguish unsafe behaviors in real time, as well as its limitations. This research not only benefits the extensive application of the method in construction safety management, but improves the effectiveness and efficiency of the method by identifying relevant future research focuses. Therefore this paper contributes to the practice as well as the body of knowledge of construction safety management.