With　the　increasing　application　of　human-robot　interaction,　collision　detection　between　robot　and　unknown　environments,　along　with　further　distinction　from　the　intentional　contact　between　human　and　robot,　have　become　urgent　problems　to　be　solved.　In　this　paper,　a　new　collision　detection　algorithm　is　proposed,　and　a　collision　distinction　method　is　further　designed　on　the　basis　of　this　algorithm.　The　generalized　momentum　and　the　convolution　method　are　used　to　develop　the　robot　convolution　filtering　dynamic　model.　Then,　the　force-sensing　observer　that　uses　only　proprioceptive　sensors　is　designed　to　observe　the　torque　deviation　of　the　joint　online　to　realize　robot　collision　detection.　At　the　same　time,　the　performance　of　the　force-sensing　observer　is　improved　by　compensating　for　joint　friction.　The　proposed　algorithm　does　not　need　any　external　sensors;　it　overcomes　the　disadvantage　of　calculation　errors　owing　to　the　acquisition　of　joint　acceleration　information.　The　filter　can　be　flexibly　selected　in　the　algorithm　according　to　the　actual　application　of　the　robot.　Moreover,　two　force-sensing　observers　are　adopted　in　the　collision　distinction　method.　The　contact　or　collision　between　a　human　and　a　robot　can　be　further　distinguished　after　setting　the　appropriate　thresholds　and　filtering　parameters.　The　collision　detection　algorithm　can　be　easily　adapted　to　different　types　of　robot　to　ensure　human　safety,　and　the　proposed　collision　distinction　method　can　be　used　to　improve　the　work　efficiency　of　the　robot.　External　force　sensing　experiments　show　that　the　low-pass　and　bandpass　observers　work　well　and　different　force　signals　can　be　observed.　The　collision　detection　and　human-robot　interaction　experiments　are　performed　to　verify　that　the　collision　detection　algorithm　and　collision　distinction　method　are　reasonable　and　effective.